Name

ccache - a fast C/C++ compiler cache

Synopsis

ccache [ccache options]
ccache [KEY=VALUE …​] compiler [compiler options]
compiler [compiler options]

The first form takes options described in Command line options below. The second form invokes the compiler, optionally using configuration options as KEY=VALUE arguments. In the third form, ccache is masquerading as the compiler as described in Run modes.

Description

Ccache is a compiler cache. It speeds up recompilation by caching the result of previous compilations and detecting when the same compilation is being done again.

Ccache has been carefully written to always produce exactly the same compiler output that you would get without the cache. The only way you should be able to tell that you are using ccache is the speed. Currently known exceptions to this goal are listed under Caveats. If you discover an undocumented case where ccache changes the output of your compiler, please let us know.

Run modes

There are two different ways to use ccache to cache a compilation:

  1. Prefix your compilation command with ccache. This method is most convenient if you just want to try out ccache or wish to use it for some specific projects. Example:

    ccache gcc -c example.c

  2. Let ccache masquerade as the compiler. This method is most useful when you wish to use ccache for all your compilations. To do this, create a symbolic link to ccache named as the compiler. For example, here is set up ccache to masquerade as gcc and g++:

    cp ccache /usr/local/bin/
ln -s ccache /usr/local/bin/gcc
ln -s ccache /usr/local/bin/g++

    On platforms that don’t support symbolic links you can simply copy ccache to the compiler name instead for a similar effect:

    cp ccache /usr/local/bin/gcc
cp ccache /usr/local/bin/g++

    And so forth. This will work as long as the directory with symbolic links or ccache copies comes before the directory with the compiler (typically /usr/bin) in PATH.

    The technique of letting ccache masquerade as the compiler works well, but currently doesn’t interact well with other tools that do the same thing. See Using ccache with other compiler wrappers.

Command line options

These command line options only apply when you invoke ccache as “ccache”. When ccache masquerades as a compiler (as described in the previous section), the normal compiler options apply and you should refer to the compiler’s documentation.

Common options

-c, --cleanup

Clean up the cache by removing not recently used cached files until the specified file number and cache size limits are not exceeded. This also recalculates the cache file count and size totals. Normally, there is no need to initiate cleanup manually as ccache keeps the cache below the specified limits at runtime and keeps statistics up to date on each compilation. Forcing a cleanup is mostly useful if you have modified the cache contents manually or believe that the cache size statistics may be inaccurate.

-C, --clear

Clear the entire cache, removing all cached files, but keeping the configuration file.

--config-path PATH

Let the command line options operate on configuration file PATH instead of the default. Using this option has the same effect as setting (overriding) the environment variable CCACHE_CONFIGPATH temporarily.

-d, --dir PATH

Let the command line options operate on cache directory PATH instead of the default. For example, to show statistics for a cache directory at /shared/ccache you can run ccache -d /shared/ccache -s. Using this option has the same effect as setting the environment variable CCACHE_DIR temporarily.

--evict-namespace NAMESPACE

Remove files created in the given namespace from the cache.

--evict-older-than AGE

Remove files used less recently than AGE from the cache. AGE should be an unsigned integer with a d (days) or s (seconds) suffix. If combined with --evict-namespace, only remove files within that namespace.

-h, --help

Print a summary of command line options.

-F NUM, --max-files NUM

Set the maximum number of files allowed in the cache to NUM. Use 0 for no limit. The value is stored in a configuration file in the cache directory and applies to all future compilations.

-M SIZE, --max-size SIZE

Set the maximum size of the files stored in the cache. SIZE should be a number followed by an optional suffix: kB, MB, GB, TB (decimal), KiB, MiB, GiB or TiB (binary). The default suffix is GiB. Use 0 for no limit. The value is stored in a configuration file in the cache directory and applies to all future compilations.

-X LEVEL, --recompress LEVEL

Recompress the cache to level LEVEL using the Zstandard algorithm. The level can be an integer, with the same semantics as the compression_level configuration option, or the special value uncompressed for no compression. See [Cache compression] for more information. This can potentially take a long time since all files in the cache need to be visited. Only files that are currently compressed with a different level than LEVEL will be recompressed.

--recompress-threads THREADS

Use up to THREADS threads when recompressing the cache. The default is to use one thread per CPU.

-o KEY=VALUE, --set-config KEY=VALUE

Set configuration option KEY to VALUE. See Configuration for more information.

-x, --show-compression

Print cache compression statistics. See Cache compression for more information. This can potentially take a long time since all files in the cache need to be visited.

-p, --show-config

Print current configuration options and from where they originate (environment variable, configuration file or compile-time default) in human-readable format.

--show-log-stats

Print statistics counters from the stats log in human-readable format. See stats_log. Use -v/--verbose once or twice for more details.

-s, --show-stats

Print a summary of configuration and statistics counters in human-readable format. Use -v/--verbose once or twice for more details.

-v, --verbose

Increase verbosity. The option can be given multiple times.

-V, --version

Print version and copyright information.

-z, --zero-stats

Zero the cache statistics (but not the configuration options).

Options for remote file-based storage

--trim-dir PATH

Remove not recently used files from directory PATH until it is at most the size specified by --trim-max-size.

Don’t use this option to trim the local cache. To trim the local cache directory to a certain size, use CCACHE_MAXSIZE=SIZE ccache -c.
--trim-max-size SIZE

Specify the maximum size for --trim-dir. SIZE should be a number followed by an optional suffix: kB, MB, GB, TB (decimal), KiB, MiB, GiB or TiB (binary). The default suffix is GiB. Use 0 for no limit.

--trim-method METHOD

Specify the method to trim a directory with --trim-dir. Possible values are:

atime

LRU (least recently used) using the file access timestamp. This is the default.

mtime

LRU (least recently used) using the file modification timestamp.

--trim-recompress LEVEL

Recompress to level LEVEL using the Zstandard algorithm when using --trim-dir. The level can be an integer, with the same semantics as the compression_level configuration option, or the special value uncompressed for no compression. See [Cache compression] for more information. This can potentially take a long time since all files in the cache need to be visited. Only files that are currently compressed with a different level than LEVEL will be recompressed.

--trim-recompress-threads THREADS

Recompress using up to THREADS threads with --trim-recompress. The default is to use one thread per CPU.

Options for scripting or debugging

--checksum-file PATH

Print the checksum (128 bit XXH3) of the file at PATH (- for standard input).

--extract-result PATH

Extract data stored in the result file at PATH (- for standard input). The data will be written to ccache-result.* files in to the current working directory. This option is only useful when debugging ccache and its behavior.

-k KEY, --get-config KEY

Print the value of configuration option KEY. See Configuration for more information.

--hash-file PATH

Print the hash (160 bit BLAKE3) of the file at PATH (- for standard input). This is only useful when debugging ccache and its behavior.

--inspect PATH

Print the content of a result or manifest file at PATH (- for standard input) to standard output in human-readable format. File content embedded in a result file will however not be printed; use --extract-result to extract the file content. This option is only useful when debugging ccache and its behavior.

--print-stats

Print statistics counter IDs and corresponding values in machine-parsable (tab-separated) format.

Extra options

When run as a compiler, ccache usually just takes the same command line options as the compiler you are using. The only exception to this is the option --ccache-skip. That option can be used to tell ccache to avoid interpreting the next option in any way and to pass it along to the compiler as-is.

--ccache-skip currently only tells ccache not to interpret the next option as a special compiler option — the option will still be included in the direct mode hash.

The reason this can be important is that ccache does need to parse the command line and determine what is an input filename and what is a compiler option, as it needs the input filename to determine the name of the resulting object file (among other things). The heuristic ccache uses when parsing the command line is that any argument that exists as a file is treated as an input file name. By using --ccache-skip you can force an option to not be treated as an input file name and instead be passed along to the compiler as a command line option.

Another case where --ccache-skip can be useful is if ccache interprets an option specially but shouldn’t, since the option has another meaning for your compiler than what ccache thinks.

Configuration

Ccache’s default behavior can be overridden by options in configuration files, which in turn can be overridden by environment variables with names starting with CCACHE_. Ccache normally reads configuration from two files: first a system-level configuration file and secondly a cache-specific configuration file. The priorities of configuration options are as follows (where 1 is highest):

  1. Command line settings in KEY=VALUE form. Example:

    ccache debug=true compiler_check="%compiler% --version" gcc -c example.c

  2. Environment variables.

  3. The cache-specific configuration file (see below).

  4. The system (read-only) configuration file <sysconfdir>/ccache.conf (typically /etc/ccache.conf or /usr/local/etc/ccache.conf).

  5. Compile-time defaults.

As a special case, if the environment variable CCACHE_CONFIGPATH is set it specifies the configuration file, and the system configuration file won’t be read.

Location of the configuration file

The location of the cache-specific configuration file is determined like this on non-Windows systems:

  1. If CCACHE_CONFIGPATH is set, use that path.

  2. Otherwise, if the environment variable CCACHE_DIR is set then use $CCACHE_DIR/ccache.conf.

  3. Otherwise, if cache_dir is set in the system configuration file then use <cache_dir>/ccache.conf.

  4. Otherwise, if there is a legacy $HOME/.ccache directory then use $HOME/.ccache/ccache.conf.

  5. Otherwise, if XDG_CONFIG_HOME is set then use $XDG_CONFIG_HOME/ccache/ccache.conf.

  6. Otherwise, use $HOME/Library/Preferences/ccache/ccache.conf (macOS) or $HOME/.config/ccache/ccache.conf (other systems).

On Windows, this is the method used to find the configuration file:

  1. If CCACHE_CONFIGPATH is set, use that path.

  2. Otherwise, if the environment variable CCACHE_DIR is set then use %CCACHE_DIR%/ccache.conf.

  3. Otherwise, if cache_dir is set in the system configuration file then use <cache_dir>\ccache.conf. The system-wide configuration on Windows is %ALLUSERSPROFILE%\ccache\ccache.conf by default. The ALLUSERSPROFILE environment variable is usually C:\ProgramData.

  4. Otherwise, if there is a legacy %USERPROFILE%\.ccache directory then use %USERPROFILE%\.ccache\ccache.conf.

  5. Otherwise, use %LOCALAPPDATA%\ccache\ccache.conf if it exists.

  6. Otherwise, use %APPDATA%\ccache\ccache.conf.

See also the cache_dir configuration option for how the cache directory location is determined.

Configuration file syntax

Configuration files are in a simple “key = value” format, one option per line. Lines starting with a hash sign are comments. Blank lines are ignored, as is whitespace surrounding keys and values. Example:

# Set maximum cache size to 10 GB:
max_size = 10G

Boolean values

Some configuration options are boolean values (i.e. truth values). In a configuration file, such values must be set to the string true or false. For the corresponding environment variables, the semantics are a bit different:

  • A set environment variable means “true” (even if set to the empty string).

  • The following case-insensitive negative values are considered an error (instead of surprising the user): 0, false, disable and no.

  • An unset environment variable means “false”.

Each boolean environment variable also has a negated form starting with CCACHE_NO. For example, CCACHE_COMPRESS can be set to force compression and CCACHE_NOCOMPRESS can be set to force no compression.

Configuration options

Below is a list of available configuration options. The corresponding environment variable name is indicated in parentheses after each configuration option key.

absolute_paths_in_stderr (CCACHE_ABSSTDERR)

This option specifies whether ccache should rewrite relative paths in the compiler’s standard error output to absolute paths. This can be useful if you use base_dir with a build system (e.g. CMake with the "Unix Makefiles" generator) that executes the compiler in a different working directory, which makes relative paths in compiler errors or warnings incorrect. The default is false.

base_dir (CCACHE_BASEDIR)

This option should be an absolute path to a directory. If set, ccache will rewrite absolute paths into paths relative to the current working directory, but only absolute paths that begin with base_dir. Cache results can then be shared for compilations in different directories even if the project uses absolute paths in the compiler command line. See also the discussion under Compiling in different directories. If set to the empty string (which is the default), no rewriting is done.

A typical path to use as base_dir is your home directory or another directory that is a parent of your project directories. Don’t use / as the base directory since that will make ccache also rewrite paths to system header files, which typically is contraproductive.

For example, say that Alice’s current working directory is /home/alice/project1/build and that she compiles like this:

ccache gcc -I/usr/include/example -I/home/alice/project2/include -c /home/alice/project1/src/example.c

Here is what ccache will actually execute for different base_dir values:

# Current working directory: /home/alice/project1/build

# With base_dir = /:
gcc -I../../../../usr/include/example -I../../project2/include -c ../src/example.c

# With base_dir = /home or /home/alice:
gcc -I/usr/include/example -I../../project2/include -c ../src/example.c

# With base_dir = /home/alice/project1 or /home/alice/project1/src:
gcc -I/usr/include/example -I/home/alice/project2/include -c ../src/example.c

If Bob has put project1 and project2 in /home/bob/stuff and both users have set base_dir to /home or /home/$USER, then Bob will get a cache hit (if they share ccache directory) since the actual command line will be identical to that of Alice:

# Current working directory: /home/bob/stuff/project1/build

# With base_dir = /home or /home/bob:
gcc -I/usr/include/example -I../../project2/include -c ../src/example.c

Without base_dir there will be a cache miss since the absolute paths will differ. With base_dir set to / there will be a cache miss since the relative path to /usr/include/example will be different. With base_dir set to /home/bob/stuff/project1 there will a cache miss since the path to project2 will be a different absolute path.

Rewriting absolute paths to relative is kind of a brittle hack. It works OK in many cases, but there might be cases where things break. One known issue is that absolute paths are not reproduced in dependency files, which can mess up dependency detection in tools like Make and Ninja. If possible, use relative paths in the first place instead instead of using base_dir.
cache_dir (CCACHE_DIR)

This option specifies where ccache will keep its cached compiler outputs.

On non-Windows systems, the default is $HOME/.ccache if such a directory exists, otherwise $XDG_CACHE_HOME/ccache if XDG_CACHE_HOME is set, otherwise $HOME/Library/Caches/ccache (macOS) or $HOME/.config/ccache (other systems).

On Windows, the default is %USERPROFILE%\.ccache if such a directory exists, otherwise %LOCALAPPDATA%\ccache.

Previous ccache versions defaulted to storing the cache in %APPDATA%\ccache on Windows. This can result in large network file transfers of the cache in domain environments and similar problems. Please check this directory for cache directories and either delete them or the whole directory, or move them to the %LOCALAPPDATA%\ccache directory.

See also Location of the configuration file.

compiler (CCACHE_COMPILER or (deprecated) CCACHE_CC)

This option can be used to force the name of the compiler to use. If set to the empty string (which is the default), ccache works it out from the command line.

compiler_check (CCACHE_COMPILERCHECK)

By default, ccache includes the modification time (“mtime”) and size of the compiler in the hash to ensure that results retrieved from the cache are accurate. If compiler plugins are used, these plugins will also be added to the hash. This option can be used to select another strategy. Possible values are:

content

Hash the content of the compiler binary. This makes ccache very slightly slower compared to mtime, but makes it cope better with compiler upgrades during a build bootstrapping process.

mtime

Hash the compiler’s mtime and size, which is fast. This is the default.

none

Don’t hash anything. This may be good for situations where you can safely use the cached results even though the compiler’s mtime or size has changed (e.g. if the compiler is built as part of your build system and the compiler’s source has not changed, or if the compiler only has changes that don’t affect code generation). You should only use none if you know what you are doing.

string:value

Hash value. This can for instance be a compiler revision number or another string that the build system generates to identify the compiler.

a command string

Hash the standard output and standard error output of the specified command. The string will be split on whitespace to find out the command and arguments to run. No other interpretation of the command string will be done, except that the special word %compiler% will be replaced with the path to the compiler. Several commands can be specified with semicolon as separator. Examples:

%compiler% -v
%compiler% -dumpmachine; %compiler% -dumpversion

You should make sure that the specified command is as fast as possible since it will be run once for each ccache invocation.

Identifying the compiler using a command is useful if you want to avoid cache misses when the compiler has been rebuilt but not changed.

Another case is when the compiler (as seen by ccache) actually isn’t the real compiler but another compiler wrapper — in that case, the default mtime method will hash the mtime and size of the other compiler wrapper, which means that ccache won’t be able to detect a compiler upgrade. Using a suitable command to identify the compiler is thus safer, but it’s also slower, so you should consider continue using the mtime method in combination with the prefix_command option if possible. See Using ccache with other compiler wrappers.

compiler_type (CCACHE_COMPILERTYPE)

Ccache normally guesses the compiler type based on the compiler name. The compiler_type option lets you force a compiler type. This can be useful if the compiler has a non-standard name but is actually one of the known compiler types. Possible values are:

auto

Guess one of the types below based on the compiler name (following symlinks). This is the default.

clang

Clang-based compiler.

clang-cl

clang-cl.

gcc

GCC-based compiler.

icl

Intel compiler on Windows.

msvc

Microsoft Visual C++ (MSVC).

nvcc

NVCC (CUDA) compiler.

other

Any compiler other than the known types.

compression (CCACHE_COMPRESS or CCACHE_NOCOMPRESS, see Boolean values above)

If true, ccache will compress data it puts in the cache. However, this option has no effect on how files are retrieved from the cache; compressed and uncompressed results will still be usable regardless of this option. The default is true.

Compression is done using the Zstandard algorithm. The algorithm is fast enough that there should be little reason to turn off compression to gain performance. One exception is if the cache is located on a compressed file system, in which case the compression performed by ccache of course is redundant.

Compression will be disabled if file cloning (the file_clone option) or hard linking (the hard_link option) is enabled.

compression_level (CCACHE_COMPRESSLEVEL)

This option determines the level at which ccache will compress object files using the real-time compression algorithm Zstandard. It only has effect if compression is enabled (which it is by default). Zstandard is extremely fast for decompression and very fast for compression for lower compression levels. The default is 0.

Semantics of compression_level:

> 0

A positive value corresponds to normal Zstandard compression levels. Lower levels (e.g. 1) mean faster compression but worse compression ratio. Higher levels (e.g. 19) mean slower compression but better compression ratio. The maximum possible value depends on the libzstd version, but at least up to 19 is available for all versions. Decompression speed is essentially the same for all levels. As a rule of thumb, use level 5 or lower since higher levels may slow down compilations noticeably. Higher levels are however useful when recompressing the cache with command line option -X/--recompress.

< 0

A negative value corresponds to Zstandard’s “ultra-fast” compression levels, which are even faster than level 1 but with less good compression ratios. For instance, level -3 corresponds to --fast=3 for the zstd command line tool. In practice, there is little use for levels lower than -5 or so.

0 (default)

The value 0 means that ccache will choose a suitable level, currently 1.

See the Zstandard documentation for more information.

cpp_extension (CCACHE_EXTENSION)

This option can be used to force a certain extension for the intermediate preprocessed file. The default is to automatically determine the extension to use for intermediate preprocessor files based on the type of file being compiled, but that sometimes doesn’t work. For example, when using the “aCC” compiler on HP-UX, set the cpp extension to i.

debug (CCACHE_DEBUG or CCACHE_NODEBUG, see Boolean values above)

If true, enable the debug mode. The debug mode creates per-object debug files that are helpful when debugging unexpected cache misses. Note however that ccache performance will be reduced slightly. See Cache debugging for more information. The default is false.

debug_dir (CCACHE_DEBUGDIR)

Specifies where to write per-object debug files if the debug mode is enabled. If set to the empty string, the files will be written next to the object file. If set to a directory, the debug files will be written with full absolute paths in that directory, creating it if needed. The default is the empty string.

For example, if debug_dir is set to /example, the current working directory is /home/user and the object file is build/output.o then the debug log will be written to /example/home/user/build/output.o.ccache-log. See also Cache debugging.

depend_mode (CCACHE_DEPEND or CCACHE_NODEPEND, see Boolean values above)

If true, the depend mode will be used. The default is false. See The depend mode.

direct_mode (CCACHE_DIRECT or CCACHE_NODIRECT, see Boolean values above)

If true, the direct mode will be used. The default is true. See The direct mode.

disable (CCACHE_DISABLE or CCACHE_NODISABLE, see Boolean values above)

When true, ccache will just call the real compiler, bypassing the cache completely. The default is false.

It is also possible to disable ccache for a specific source code file by adding the string ccache:disable in a comment in the first 4096 bytes of the file.

extra_files_to_hash (CCACHE_EXTRAFILES)

This option is a list of paths to files that ccache will include in the the hash sum that identifies the build. The list separator is semicolon on Windows systems and colon on other systems.

file_clone (CCACHE_FILECLONE or CCACHE_NOFILECLONE, see Boolean values above)

If true, ccache will attempt to use file cloning (also known as “copy on write”, “CoW” or “reflinks”) to store and fetch cached compiler results. file_clone has priority over hard_link. The default is false.

Files stored by cloning cannot be compressed, so the cache size will likely be significantly larger if this option is enabled. However, performance may be improved depending on the use case.

Unlike the hard_link option, file_clone is completely safe to use, but not all file systems support the feature. For such file systems, ccache will fall back to use plain copying (or hard links if hard_link is enabled).

hash_dir (CCACHE_HASHDIR or CCACHE_NOHASHDIR, see Boolean values above)

If true (which is the default), ccache will include the current working directory (CWD) in the hash that is used to distinguish two compilations when generating debug info (compiler option -g with variations). Exception: The CWD will not be included in the hash if base_dir is set (and matches the CWD) and the compiler option -fdebug-prefix-map is used. See also the discussion under Compiling in different directories.

The reason for including the CWD in the hash by default is to prevent a problem with the storage of the current working directory in the debug info of an object file, which can lead ccache to return a cached object file that has the working directory in the debug info set incorrectly.

You can disable this option to get cache hits when compiling the same source code in different directories if you don’t mind that CWD in the debug info might be incorrect.

ignore_headers_in_manifest (CCACHE_IGNOREHEADERS)

This option is a list of paths to files (or directories with headers) that ccache will not include in the manifest list that makes up the direct mode. Note that this can cause stale cache hits if those headers do indeed change. The list separator is semicolon on Windows systems and colon on other systems.

ignore_options (CCACHE_IGNOREOPTIONS)

This option is a space-delimited list of compiler options that ccache will exclude from the hash. Excluding a compiler option from the hash can be useful when you know it doesn’t affect the result (but ccache doesn’t know that), or when it does and you don’t care. If a compiler option in the list is suffixed with an asterisk (*) it will be matched as a prefix. For example, -fmessage-length=* will match both -fmessage-length=20 and -fmessage-length=70.

inode_cache (CCACHE_INODECACHE or CCACHE_NOINODECACHE, see Boolean values above)

If true, ccache will cache source file hashes based on device, inode and timestamps. This reduces the time spent on hashing include files since the result can be reused between compilations. The default is true. The feature requires temporary_dir to be located on a local filesystem of a supported type.

The inode cache feature is currently not available on Windows.
keep_comments_cpp (CCACHE_COMMENTS or CCACHE_NOCOMMENTS, see Boolean values above)

If true, ccache will not discard the comments before hashing preprocessor output. The default is false. This can be used to check documentation with -Wdocumentation.

log_file (CCACHE_LOGFILE)

If set to a file path, ccache will write information on what it is doing to the specified file. This is useful for tracking down problems.

If set to syslog, ccache will log using syslog() instead of to a file. If you use rsyslogd, you can add something like this to /etc/rsyslog.conf or a file in /etc/rsyslog.d:

# log ccache to file
:programname, isequal, "ccache"         /var/log/ccache
# remove from syslog
& ~
max_files (CCACHE_MAXFILES)

This option specifies the maximum number of files to keep in the cache. Use 0 for no limit (which is the default). See also Cache size management.

max_size (CCACHE_MAXSIZE)

This option specifies the maximum size of the cache. Use 0 for no limit. The default value is 5G. Available suffixes: k, M, G, T (decimal) and Ki, Mi, Gi, Ti (binary). The default suffix is G. See also Cache size management.

msvc_dep_prefix (CCACHE_MSVC_DEP_PREFIX)

This option specifies the prefix of included files output for MSVC compiler. The default prefix is “Note: including file:”. If you use a localized compiler, this should be set accordingly.

namespace (CCACHE_NAMESPACE)

If set, the namespace string will be added to the hashed data for each compilation. This will make the associated cache entries logically separate from cache entries with other namespaces, but they will still share the same storage space. Cache entries can also be selectively removed from the local cache with the command line option --evict-namespace, potentially in combination with --evict-older-than.

For instance, if you use the same local cache for several disparate projects, you can use a unique namespace string for each one. This allows you to remove cache entries that belong to a certain project if you stop working with that project.

path (CCACHE_PATH)

If set, ccache will search directories in this list when looking for the real compiler. The list separator is semicolon on Windows systems and colon on other systems. If not set, ccache will look for the first executable matching the compiler name in the normal PATH that isn’t a symbolic link to ccache itself.

pch_external_checksum (CCACHE_PCH_EXTSUM or CCACHE_NOPCH_EXTSUM, see Boolean values above)

When this option is set, and ccache finds a precompiled header file, ccache will look for a file with the extension “.sum” added (e.g. “pre.h.gch.sum”), and if found, it will hash this file instead of the precompiled header itself to work around the performance penalty of hashing very large files.

prefix_command (CCACHE_PREFIX)

This option adds a list of prefixes (separated by space) to the command line that ccache uses when invoking the compiler. See also Using ccache with other compiler wrappers.

prefix_command_cpp (CCACHE_PREFIX_CPP)

This option adds a list of prefixes (separated by space) to the command line that ccache uses when invoking the preprocessor.

read_only (CCACHE_READONLY or CCACHE_NOREADONLY, see Boolean values above)

If true, ccache will attempt to use existing cached results, but it will not add new results to any cache backend. Statistics counters will still be updated, though, unless the stats option is set to false.

If you are using this because your ccache directory is read-only, you need to set temporary_dir since ccache will fail to create temporary files otherwise. You may also want to set stats to false make ccache not even try to update stats files.

read_only_direct (CCACHE_READONLY_DIRECT or CCACHE_NOREADONLY_DIRECT, see Boolean values above)

Just like read_only except that ccache will only try to retrieve results from the cache using the direct mode, not the preprocessor mode. See documentation for read_only regarding using a read-only ccache directory.

recache (CCACHE_RECACHE or CCACHE_NORECACHE, see Boolean values above)

If true, ccache will not use any previously stored result. New results will still be cached, possibly overwriting any pre-existing results.

remote_only (CCACHE_REMOTE_ONLY or CCACHE_NOREMOTE_ONLY, see Boolean values above)

If true, ccache will only use remote storage. The default is false. Note that cache statistics counters will still be kept in the local cache directory unless stats is false. See also Storage interaction.

remote_storage (CCACHE_REMOTE_STORAGE)

This option specifies one or several storage backends (separated by space) to query after checking the local cache (unless remote_only is true). See Remote storage backends for documentation of syntax and available backends.

Examples:

  • file:/shared/nfs/directory

  • file:///shared/nfs/one|read-only file:///shared/nfs/two

  • file:///Z:/example/windows/folder

  • http://example.com/cache

  • redis://example.com

    In previous ccache versions this option was called secondary_storage (CCACHE_SECONDARY_STORAGE), which can still be used as an alias.
reshare (CCACHE_RESHARE or CCACHE_NORESHARE, see Boolean values above)

If true, ccache will write results to remote storage even for local storage cache hits. The default is false.

run_second_cpp (CCACHE_CPP2 or CCACHE_NOCPP2, see Boolean values above)

If true, ccache will first run the preprocessor to preprocess the source code (see The preprocessor mode) and then on a cache miss run the compiler on the source code to get hold of the object file. This is the default.

If false, ccache will first run preprocessor to preprocess the source code and then on a cache miss run the compiler on the preprocessed source code instead of the original source code. This makes cache misses slightly faster since the source code only has to be preprocessed once. The downside is that some compilers won’t produce the same result (for instance diagnostics warnings) when compiling preprocessed source code.

A solution to the above mentioned downside is to set run_second_cpp to false and pass -fdirectives-only (for GCC) or -frewrite-includes (for Clang) to the compiler. This will cause the compiler to leave the macros and other preprocessor information, and only process the #include directives. When run in this way, the preprocessor arguments will be passed to the compiler since it still has to do some preprocessing (like macros).

This option is ignored with MSVC, as there is no way to make it compile without preprocessing first.

sloppiness (CCACHE_SLOPPINESS)

By default, ccache tries to give as few false cache hits as possible. However, in certain situations it’s possible that you know things that ccache can’t take for granted. This option makes it possible to tell ccache to relax some checks in order to increase the hit rate. The value should be a comma-separated string with one or several of the following values:

clang_index_store

Ignore the Clang compiler option -index-store-path and its argument when computing the manifest hash. This is useful if you use Xcode, which uses an index store path derived from the local project path. Note that the index store won’t be updated correctly on cache hits if you enable this sloppiness.

file_stat_matches

Ccache normally examines a file’s contents to determine whether it matches the cached version. With this sloppiness set, ccache will consider a file as matching its cached version if the mtimes and ctimes match.

file_stat_matches_ctime

Ignore ctimes when file_stat_matches is enabled. This can be useful when backdating files' mtimes in a controlled way.

gcno_cwd

By default, ccache will include the current working directory in the hash when producing a .gcno file (when compiling with -ftest-coverage or --coverage). This is because GCC 9+ includes the current working directory in the .gcno file. The gcno_cwd sloppiness makes ccache not hash the current working directory so that you can get cache hits when compiling in different directories, with the tradeoff of potentially getting an incorrect directory in the .gcno file. gcno_cwd also disables hashing of the current working directory if -fprofile-abs-path is used.

include_file_ctime

By default, ccache will disable the direct mode if an include file has too new ctime. This sloppiness disables that check. See also [Handling of newly created header files].

include_file_mtime

By default, ccache will disable the direct mode if an include file has too new mtime. This sloppiness disables that check. See also [Handling of newly created header files].

ivfsoverlay

Ignore the Clang compiler option -ivfsoverlay and its argument. This is useful if you use Xcode, which uses a virtual file system (VFS) for things like combining Objective-C and Swift code.

locale

Ccache includes the environment variables LANG, LC_ALL, LC_CTYPE and LC_MESSAGES in the hash by default since they may affect localization of compiler warning messages. Set this sloppiness to tell ccache not to do that.

modules

By default, ccache will not cache compilations if -fmodules is used since it cannot hash the state of compiler’s internal representation of relevant modules. This sloppiness allows caching in such a case. See C++ modules for more information.

pch_defines

Be sloppy about #define directives when precompiling a header file. See Precompiled headers for more information.

random_seed

Ignore the -frandom-seed option and its arguments when computing the input hash. This is useful if your build system generates different seeds between builds and you are OK with reusing cached results.

system_headers

By default, ccache will also include all system headers in the manifest. With this sloppiness set, ccache will only include system headers in the hash but not add the system header files to the list of include files.

time_macros

Ignore __DATE__, __TIME__ and __TIMESTAMP__ being present in the source code.

See the discussion under Troubleshooting for more information.

stats (CCACHE_STATS or CCACHE_NOSTATS, see Boolean values above)

If true, ccache will update the statistics counters on each compilation. The default is true. If false, [automatic cleanup] will be disabled as well.

stats_log (CCACHE_STATSLOG)

If set to a file path, ccache will write statistics counter updates to the specified file. This is useful for getting statistics for individual builds. To show a summary of the current stats log, use ccache --show-log-stats.

Lines in the stats log starting with a hash sign (#) are comments.
temporary_dir (CCACHE_TEMPDIR)

This option specifies where ccache will put temporary files. The default is $XDG_RUNTIME_DIR/ccache-tmp (typically /run/user/<UID>/ccache-tmp) if XDG_RUNTIME_DIR is set and the directory exists, otherwise <cache_dir>/tmp.

In previous versions of ccache, CCACHE_TEMPDIR had to be on the same filesystem as the CCACHE_DIR path, but this requirement has been relaxed.
umask (CCACHE_UMASK)

This option (an octal integer) specifies the umask for files and directories in the cache directory. This is mostly useful when you wish to share your cache with other users.

Disabling ccache

To disable ccache completely for all invocations, set disable = true (CCACHE_DISABLE=1). You can also disable ccache for a certain source code file by adding the string ccache:disable in a comment in the first 4096 bytes of the file. In the latter case the Ccache disabled statistics counter will be increased.

Remote storage backends

The remote_storage option lets you configure ccache to use one or several remote storage backends. By default, the local cache directory located in cache_dir will be queried first and remote storage second, but remote_only can be set to true to disable local storage. Note that cache statistics counters will still be kept in the local cache directory — remote storage backends only store compilation results and manifests.

A remote storage backend is specified with a URL, optionally followed by a pipe (|) and a pipe-separated list of attributes. An attribute is key=value or just key as a short form of key=true. Attribute values must be percent-encoded if they contain percent, pipe or space characters.

Attributes for all backends

These optional attributes are available for all remote storage backends:

  • read-only: If true, only read from this backend, don’t write. The default is false.

  • shards: A comma-separated list of names for sharding (partitioning) the cache entries using Rendezvous hashing, typically to spread the cache over a server cluster. When set, the storage URL must contain an asterisk (*), which will be replaced by one of the shard names to form a real URL. A shard name can optionally have an appended weight within parentheses to indicate how much of the key space should be associated with that shard. A shard with weight w will contain w/S of the cache, where S is the sum of all shard weights. A weight could for instance be set to represent the available memory for a memory cache on a specific server. The default weight is 1.

    Examples:

    • redis://cache-*.example.com|shards=a(3),b(1),c(1.5) will put 55% (3/5.5) of the cache on redis://cache-a.example.com, 18% (1/5.5) on redis://cache-b.example.com and 27% (1.5/5.5) on redis://cache-c.example.com.

    • http://example.com/*|shards=alpha,beta will put 50% of the cache on http://example.com/alpha and 50% on http://example.com/beta.

Storage interaction

The table below describes the interaction between local and remote storage on cache hits and misses if remote_only is false (which is the default):

Local storage Remote storage What happens

miss

miss

Compile, write to local, write to remote[1]

miss

hit

Read from remote, write to local

hit

-

Read from local, don’t write to remote[2]

[1] Unless remote storage has attribute read-only=true.
[2] Unless local storage is set to share its cache hits with the reshare option.

If remote_only is true:

Local storage Remote storage What happens

-

miss

Compile, write to remote, don’t write to local

-

hit

Read from remote, don’t write to local

File storage backend

URL format: file:DIRECTORY or file://[HOST]DIRECTORY

This backend stores data as separate files in a directory structure below DIRECTORY, similar (but not identical) to the local cache storage. A typical use case for this backend would be sharing a cache on an NFS directory. DIRECTORY must start with a slash. HOST can be the empty string or localhost. On Windows, HOST can also be the name of a server hosting a shared folder.

ccache will not perform any cleanup of the storage — that has to be done by other means, for instance by running ccache --trim-dir periodically.

Examples:

  • file:/shared/nfs/directory

  • file:///shared/nfs/directory|umask=002|update-mtime=true

  • file:///Z:/example/windows/folder

  • file://example.com/shared/ccache%20folder

Optional attributes:

  • layout: How to store file under the cache directory. Available values:

    • flat: Store all files directly under the cache directory.

    • subdirs: Store files in 256 subdirectories of the cache directory.

    The default is subdirs.

  • umask: This attribute (an octal integer) overrides the umask to use for files and directories in the cache directory.

  • update-mtime: If true, update the modification time (mtime) of cache entries that are read. The default is false.

HTTP storage backend

URL format: http://HOST[:PORT][/PATH]

This backend stores data in an HTTP-compatible server. The required HTTP methods are GET, PUT and DELETE.

ccache will not perform any cleanup of the storage — that has to be done by other means, for instance by running ccache --trim-dir periodically.
HTTPS is not supported.
See How to set up HTTP storage for hints on how to set up an HTTP server for use with ccache.

Examples:

  • http://localhost

  • http://someusername:p4ssw0rd@example.com/cache/

  • http://localhost:8080|layout=bazel|connect-timeout=50

Optional attributes:

  • bearer-token: Bearer token used to authorize the HTTP requests.

  • connect-timeout: Timeout (in ms) for network connection. The default is 100.

  • keep-alive: If true, keep the HTTP connection to the storage server open to avoid reconnects. The default is true.

  • layout: How to map key names to the path part of the URL. Available values:

    • bazel: Store values in a format compatible with the Bazel HTTP caching protocol. More specifically, the entries will be stored as 64 hex digits under the /ac/ part of the cache.

      You may have to disable verification of action cache values in the server for this to work since ccache entries are not valid action result metadata values.

    • flat: Append the key directly to the path part of the URL (with a leading slash if needed).

    • subdirs: Append the first two characters of the key to the URL (with a leading slash if needed), followed by a slash and the rest of the key. This divides the entries into 256 buckets.

    The default is subdirs.

  • operation-timeout: Timeout (in ms) for HTTP requests. The default is 10000.

Redis storage backend

URL formats:

redis://[[USERNAME:]PASSWORD@]HOST[:PORT][/DBNUMBER]
redis+unix:SOCKET_PATH[?db=DBNUMBER]
redis+unix://[[USERNAME:]PASSWORD@localhost]SOCKET_PATH[?db=DBNUMBER]

This backend stores data in a Redis (or Redis-compatible) server. There are implementations for both memory-based and disk-based storage. PORT defaults to 6379 and DBNUMBER defaults to 0.

ccache will not perform any cleanup of the Redis storage, but you can configure LRU eviction.
See How to set up Redis storage for hints on setting up a Redis server for use with ccache.
You can set up a cluster of Redis servers using the shards attribute described in Remote storage backends.

Examples:

  • redis://localhost

  • redis://p4ssw0rd@cache.example.com:6379/0|connect-timeout=50

  • redis+unix:/run/redis.sock

  • redis+unix:///run/redis.sock

  • redis+unix://p4ssw0rd@localhost/run/redis.sock?db=0

Optional attributes:

  • connect-timeout: Timeout (in ms) for network connection. The default is 100.

  • operation-timeout: Timeout (in ms) for Redis commands. The default is 10000.

Cache size management

By default, ccache has a 5 GB limit on the total size of files in the cache and no limit on the number of files. You can set different limits using the command line options -M/--max-size and -F/--max-files. Use the -s/--show-stats option to see the cache size and the currently configured limits (in addition to other various statistics).

Cleanup can be triggered in two different ways: automatic and manual.

Automatic cleanup

After a new compilation result has been written to the local cache, ccache will trigger an automatic cleanup if max_size or max_files is exceeded. The cleanup removes cache entries in LRU (least recently used) order based on the modification time (mtime) of files in the cache. For this reason, ccache updates mtime of the cache files read on a cache hit to mark them as recently used.

Manual cleanup

You can run ccache -c/--cleanup to force cleanup of the whole cache. This will recalculate the cache size information and also make sure that the cache size does not exceed max_size and max_files.

Cache compression

Ccache will by default compress all data it puts into the cache using the compression algorithm Zstandard (zstd) using compression level 1. The algorithm is fast enough that there should be little reason to turn off compression to gain performance. One exception is if the cache is located on a compressed file system, in which case the compression performed by ccache of course is redundant. See the documentation for the configuration options compression and compression_level for more information.

You can use the command line option -x/--show-compression to print information related to compression. Example:

Total data:           14.8 GB (16.0 GB disk blocks)
Compressed data:      11.3 GB (30.6% of original size)
  Original size:      36.9 GB
  Compression ratio: 3.267 x  (69.4% space savings)
Incompressible data:   3.5 GB

Notes:

  • The “disk blocks” size is the cache size when taking disk block size into account. This value should match the “Cache size” value from “ccache --show-stats”. The other size numbers refer to actual content sizes.

  • “Compressed data” refers to result and manifest files stored in the cache.

  • “Incompressible data” refers to files that are always stored uncompressed (triggered by enabling file_clone or hard_link) or unknown files (for instance files created by older ccache versions).

  • The compression ratio is affected by compression_level.

The cache data can also be recompressed to another compression level (or made uncompressed) with the command line option -X/--recompress. If you choose to disable compression by default or to use a low compression level, you can (re)compress newly cached data with a higher compression level after the build or at another time when there are more CPU cycles available, for instance every night. Full recompression potentially takes a lot of time, but only files that are currently compressed with a different level than the target level will be recompressed.

Cache statistics

ccache --show-stats shows a summary of statistics, including cache size, cleanups (number of performed cleanups, either implicitly due to a cache size limit being reached or due to explicit ccache -c calls), overall hit rate, hit rate for direct/preprocessed modes and hit rate for local and remote storage.

The summary also includes counters called “Errors” and “Uncacheable”, which are sums of more detailed counters. To see those detailed counters, use the -v/--verbose flag. The verbose mode can show the following counters:

Counter Description

Autoconf compile/link

Uncacheable compilation or linking by an Autoconf test.

Bad compiler arguments

Malformed compiler argument, e.g. missing a value for a compiler option that requires an argument or failure to read a file specified by a compiler option argument.

Called for linking

The compiler was called for linking, not compiling. Ccache only supports compilation of a single file, i.e. calling the compiler with the -c option to produce a single object file from a single source file.

Called for preprocessing

The compiler was called for preprocessing, not compiling.

Ccache disabled

Ccache was disabled by a ccache:disable string in the source code file.

Could not use modules

Preconditions for using C++ modules were not fulfilled.

Could not use precompiled header

Preconditions for using precompiled headers were not fulfilled.

Could not write to output file

The output path specified with -o could not be written to.

Compilation failed

The compilation failed. No result stored in the cache.

Compiler check failed

A compiler check program specified by compiler_check (CCACHE_COMPILERCHECK) failed.

Compiler output file missing

One of the files expected to be produced by the compiler was missing after compilation.

Compiler produced empty output

The compiler’s output file (typically an object file) was empty after compilation.

Could not find the compiler

The compiler to execute could not be found.

Error hashing extra file

Failure reading a file specified by extra_files_to_hash (CCACHE_EXTRAFILES).

Forced recache

CCACHE_RECACHE was used to overwrite an existing result.

Internal error

Unexpected failure, e.g. due to problems reading/writing the cache.

Missing cache file

A file was unexpectedly missing from the cache. This only happens in rare situations, e.g. if one ccache instance is about to get a file from the cache while another instance removed the file as part of cache cleanup.

Multiple source files

The compiler was called to compile multiple source files in one go. This is not supported by ccache.

No input file

No input file was specified to the compiler.

Output to stdout

The compiler was instructed to write its output to standard output using -o -. This is not supported by ccache.

Preprocessing failed

Preprocessing the source code using the compiler’s -E option failed.

Unsupported code directive

Code like the assembler .incbin directive was found. This is not supported by ccache.

Unsupported compiler option

A compiler option not supported by ccache was found.

Unsupported environment variable

An environment variable not supported by ccache was set.

Unsupported source language

A source language e.g. specified with -x was unsupported by ccache.

How ccache works

The basic idea is to detect when you are compiling exactly the same code a second time and reuse the previously produced output. The detection is done by hashing different kinds of information that should be unique for the compilation and then using the hash sum to identify the cached output. Ccache uses BLAKE3, a very fast cryptographic hash algorithm, for the hashing. On a cache hit, ccache is able to supply all of the correct compiler outputs (including all warnings, dependency file, etc) from the cache. Data stored in the cache is checksummed with XXH3, an extremely fast non-cryptographic algorithm, to detect corruption.

Ccache has two ways of gathering information used to look up results in the cache:

  • the preprocessor mode, where ccache runs the preprocessor on the source code and hashes the result

  • the direct mode, where ccache hashes the source code and include files directly

The direct mode is generally faster since running the preprocessor has some overhead.

If no previous result is detected (i.e., there is a cache miss) using the direct mode, ccache will fall back to the preprocessor mode unless the depend mode is enabled. In the depend mode, ccache never runs the preprocessor, not even on cache misses. Read more in The depend mode below.

Common hashed information

The following information is always included in the hash:

  • the extension used by the compiler for a file with preprocessor output (normally .i for C code and .ii for C++ code)

  • the compiler’s size and modification time (or other compiler-specific information specified by compiler_check)

  • the name of the compiler

  • the current directory (if hash_dir is enabled)

  • contents of files specified by extra_files_to_hash (if any)

The preprocessor mode

In the preprocessor mode, the hash is formed of the common information and:

  • the preprocessor output from running the compiler with -E

  • the command line options except those that affect include files (-I, -include, -D, etc; the theory is that these command line options will change the preprocessor output if they have any effect at all)

  • any standard error output generated by the preprocessor

Based on the hash, the cached compilation result can be looked up directly in the cache.

The direct mode

In the direct mode, the hash is formed of the common information and:

  • the input source file

  • the compiler options

Based on the hash, a data structure called “manifest” is looked up in the cache. The manifest contains:

  • references to cached compilation results (object file, dependency file, etc) that were produced by previous compilations that matched the hash

  • paths to the include files that were read at the time the compilation results were stored in the cache

  • hash sums of the include files at the time the compilation results were stored in the cache

The current contents of the include files are then hashed and compared to the information in the manifest. If there is a match, ccache knows the result of the compilation. If there is no match, ccache falls back to running the preprocessor. The output from the preprocessor is parsed to find the include files that were read. The paths and hash sums of those include files are then stored in the manifest along with information about the produced compilation result.

There is a catch with the direct mode: header files that were used by the compiler are recorded, but header files that were not used, but would have been used if they existed, are not. So, when ccache checks if a result can be taken from the cache, it currently can’t check if the existence of a new header file should invalidate the result. In practice, the direct mode is safe to use in the absolute majority of cases.

The direct mode will be disabled if any of the following holds:

  • direct_mode is false

  • a modification time of one of the include files is too new (needed to avoid a race condition)

  • a compiler option not supported by the direct mode is used:

    • a -Wp,* compiler option other than -Wp,-MD,<path>, -Wp,-MMD,<path>, -Wp,-D<macro[=defn]> or -Wp,-U<macro>

    • -Xpreprocessor

  • the string __TIME__ is present in the source code

The depend mode

If the depend mode is enabled, ccache will not use the preprocessor at all. The hash used to identify results in the cache will be based on the direct mode hash described above plus information about include files read from the dependency list generated by MSVC with /showIncludes, or the dependency file generated by other compilers with -MD or -MMD.

Advantages:

  • The ccache overhead of a cache miss will be much smaller.

  • Not running the preprocessor at all can be good if compilation is performed remotely, for instance when using distcc or similar; ccache then won’t make potentially costly preprocessor calls on the local machine.

Disadvantages:

  • The cache hit rate will likely be lower since any change to compiler options or source code will make the hash different. Compare this with the default setup where ccache will fall back to the preprocessor mode, which is tolerant to some types of changes of compiler options and source code changes.

  • If -MD is used, the manifest entries will include system header files as well, thus slowing down cache hits slightly, just as using -MD slows down make. This is also the case for MSVC with /showIncludes.

  • If -MMD is used, the manifest entries will not include system header files, which means ccache will ignore changes in them.

The depend mode will be disabled if any of the following holds:

  • depend_mode is false.

  • run_second_cpp is false.

  • The compiler is not generating dependencies using -MD or -MMD (for MSVC, /showIncludes is added automatically if not specified by the user).

Handling of newly created header files

If modification time (mtime) or status change time (ctime) of one of the include files is equal to (or newer than) the time compilation is being done, ccache disables the direct mode (or, in the case of a precompiled header, disables caching completely). This done as a safety measure to avoid a race condition (see below).

To be able to use a newly created header files in direct mode (or use a newly precompiled header), either:

  • create the include file earlier in the build process, or

  • set sloppiness to include_file_ctime,include_file_mtime if you are willing to take the risk, for instance if you know that your build system is robust enough not to trigger the race condition.

For reference, the race condition mentioned above consists of these events:

  1. The preprocessor is run.

  2. An include file is modified by someone.

  3. The new include file is hashed by ccache.

  4. The real compiler is run on the preprocessor’s output, which contains data from the old header file.

  5. The wrong object file is stored in the cache.

Cache debugging

To find out what information ccache actually is hashing, you can enable the debug mode via the configuration option debug or by setting CCACHE_DEBUG in the environment. This can be useful if you are investigating why you don’t get cache hits. Note that performance will be reduced slightly.

When the debug mode is enabled, ccache will create up to five additional files next to the object file:

Filename Description

<objectfile>.<timestamp>.ccache-input-c

Binary input hashed by both the direct mode and the preprocessor mode.

<objectfile>.<timestamp>.ccache-input-d

Binary input only hashed by the direct mode.

<objectfile>.<timestamp>.ccache-input-p

Binary input only hashed by the preprocessor mode.

<objectfile>.<timestamp>.ccache-input-text

Human-readable combined diffable text version of the three files above.

<objectfile>.<timestamp>.ccache-log

Log for this object file.

 
 The timestamp format is
`<year><month><day>_<hour><minute><second>_<microsecond>`.

If debug_dir (environment variable CCACHE_DEBUGDIR) is set, the files above will be written to that directory with full absolute paths instead of next to the object file.

In the direct mode, ccache uses the 160 bit BLAKE3 hash of the “ccache-input-c” + “ccache-input-d” data (where + means concatenation), while the “ccache-input-c” + “ccache-input-p” data is used in the preprocessor mode.

The “ccache-input-text” file is a combined text version of the three binary input files. It has three sections (“COMMON”, “DIRECT MODE” and “PREPROCESSOR MODE”), which is turn contain annotations that say what kind of data comes next.

To debug why you don’t get an expected cache hit for an object file, you can do something like this:

  1. Enable debug (CCACHE_DEBUG).

  2. Build.

  3. Clean and build again.

  4. Compare the <objectfile>.<timestamp>.ccache-input-text files for the two builds. This together with the <objectfile>.<timestamp>.ccache-log files should give you some clues about what is happening.

Compiling in different directories

Some information included in the hash that identifies a unique compilation can contain absolute paths:

  • The preprocessed source code may contain absolute paths to include files if the compiler option -g is used or if absolute paths are given to -I and similar compiler options.

  • Paths specified by compiler options (such as -I, -MF, etc) on the command line may be absolute.

  • The source code file path may be absolute, and that path may substituted for __FILE__ macros in the source code or included in warnings emitted to standard error by the preprocessor.

This means that if you compile the same code in different locations, you can’t share compilation results between the different build directories since you get cache misses because of the absolute build directory paths that are part of the hash.

Here’s what can be done to enable cache hits between different build directories:

  • If you build with -g (or similar) to add debug information to the object file, you must either:

    • use the compiler option -fdebug-prefix-map=<old>=<new> for relocating debug info to a common prefix (e.g. -fdebug-prefix-map=$PWD=.); or

    • set hash_dir = false.

  • If you use absolute paths anywhere on the command line (e.g. the source code file path or an argument to compiler options like -I and -MF), you must set base_dir to an absolute path to a “base directory”. Ccache will then rewrite absolute paths under that directory to relative before computing the hash.

Precompiled headers

Ccache has support for precompiled headers with GCC and Clang. However, you have to do some things to make it work properly:

  • You must set sloppiness to pch_defines,time_macros. The reason is that ccache can’t tell whether __TIME__, __DATE__ or __TIMESTAMP__ is used when using a precompiled header. Further, it can’t detect changes in #defines in the source code because of how preprocessing works in combination with precompiled headers.

  • You may also want to include include_file_mtime,include_file_ctime in sloppiness. See Handling of newly created header files.

  • You must either:

    • use the compiler option -include to include the precompiled header (i.e., don’t use #include in the source code to include the header; the filename itself must be sufficient to find the header, i.e. -I paths are not searched); or

    • (for the Clang compiler) use the compiler option -include-pch to include the PCH file generated from the precompiled header; or

    • (for the GCC compiler) add the compiler option -fpch-preprocess when compiling.

  • If you use Clang, you must compile with -fno-pch-timestamp.

If you don’t do this, either the non-precompiled version of the header file will be used (if available) or ccache will fall back to running the real compiler and increase the statistics counter “Preprocessing failed” (if the non-precompiled header file is not available).

C++ modules

Ccache has support for Clang’s -fmodules option. In practice ccache only additionally hashes module.modulemap files; it does not know how Clang handles its cached binary form of modules so those are ignored. This should not matter in practice: as long as everything else (including module.modulemap files) is the same the cached result should work. Still, you must set sloppiness to modules to allow caching.

You must use both direct mode and depend mode. When using the preprocessor mode Clang does not provide enough information to allow hashing of module.modulemap files.

Sharing a local cache

A group of developers can increase the cache hit rate by sharing a local cache directory. To share a local cache without unpleasant side effects, the following conditions should to be met:

  • Use the same cache directory.

  • Make sure that the configuration option hard_link is false (which is the default).

  • Make sure that all users are in the same group.

  • Set the configuration option umask to 002. This ensures that cached files are accessible to everyone in the group.

  • Make sure that all users have write permission in the entire cache directory (and that you trust all users of the shared cache).

  • Make sure that the setgid bit is set on all directories in the cache. This tells the filesystem to inherit group ownership for new directories. The following command might be useful for this:

    find $CCACHE_DIR -type d | xargs chmod g+s

The reason to avoid the hard link mode is that the hard links cause unwanted side effects, as all links to a cached file share the file’s modification timestamp. This results in false dependencies to be triggered by timestamp-based build systems whenever another user links to an existing file. Typically, users will see that their libraries and binaries are relinked without reason.

You may also want to make sure that a base directory is set appropriately, as discussed in a previous section.

Sharing a cache on NFS

It is possible to put the cache directory on an NFS filesystem (or similar filesystems), but keep in mind that:

  • Having the cache on NFS may slow down compilation. Make sure to do some benchmarking to see if it’s worth it.

  • Ccache hasn’t been tested very thoroughly on NFS.

A tip is to set temporary_dir to a directory on the local host to avoid NFS traffic for temporary files.

It is recommended to use the same operating system version when using a shared cache. If operating system versions are different then system include files will likely be different and there will be few or no cache hits between the systems. One way of improving cache hit rate in that case is to set sloppiness to system_headers to ignore system headers.

An alternative to putting the main cache directory on NFS is to set up a remote storage file cache.

Using ccache with other compiler wrappers

The recommended way of combining ccache with another compiler wrapper (such as “distcc”) is by letting ccache execute the compiler wrapper. This is accomplished by defining prefix_command, for example by setting the environment variable CCACHE_PREFIX to the name of the wrapper (e.g. distcc). Ccache will then prefix the command line with the specified command when running the compiler. To specify several prefix commands, set prefix_command to a colon-separated list of commands.

Unless you set compiler_check to a suitable command (see the description of that configuration option), it is not recommended to use the form ccache anotherwrapper compiler args as the compilation command. It’s also not recommended to use the masquerading technique for the other compiler wrapper. The reason is that by default, ccache will in both cases hash the mtime and size of the other wrapper instead of the real compiler, which means that:

  • Compiler upgrades will not be detected properly.

  • The cached results will not be shared between compilations with and without the other wrapper.

Another minor thing is that if prefix_command is used, ccache will not invoke the other wrapper when running the preprocessor, which increases performance. You can use prefix_command_cpp if you also want to invoke the other wrapper when doing preprocessing (normally by adding -E).

Caveats

  • The direct mode fails to pick up new header files in some rare scenarios. See The direct mode above.

Troubleshooting

General

A general tip for getting information about what ccache is doing is to enable debug logging by setting the configuration option debug (or the environment variable CCACHE_DEBUG); see Cache debugging for more information. Another way of keeping track of what is happening is to check the output of ccache -s.

Performance

Ccache has been written to perform well out of the box, but sometimes you may have to do some adjustments of how you use the compiler and ccache in order to improve performance.

Since ccache works best when I/O is fast, put the cache directory on a fast storage device if possible. Having lots of free memory so that files in the cache directory stay in the disk cache is also preferable.

A good way of monitoring how well ccache works is to run ccache -s before and after your build and then compare the statistics counters. Here are some common problems and what may be done to increase the hit rate:

  • If the counter for preprocessed cache hits has been incremented instead of the one for direct cache hits, ccache has fallen back to preprocessor mode, which is generally slower. Some possible reasons are:

    • The source code has been modified in such a way that the preprocessor output is not affected.

    • Compiler arguments that are hashed in the direct mode but not in the preprocessor mode have changed (-I, -include, -D, etc) and they didn’t affect the preprocessor output.

    • The compiler option -Xpreprocessor or -Wp,* (except -Wp,-MD,<path>, -Wp,-MMD,<path>, and -Wp,-D<define>) is used.

    • This was the first compilation with a new value of the base directory.

    • A modification or status change time of one of the include files is too new (created the same second as the compilation is being done). See Handling of newly created header files.

    • The __TIME__ preprocessor macro is (potentially) being used. Ccache turns off direct mode if __TIME__ is present in the source code. This is done as a safety measure since the string indicates that a __TIME__ macro may affect the output. (To be sure, ccache would have to run the preprocessor, but the sole point of the direct mode is to avoid that.) If you know that __TIME__ isn’t used in practise, or don’t care if ccache produces objects where __TIME__ is expanded to something in the past, you can set sloppiness to time_macros.

    • The __DATE__ preprocessor macro is (potentially) being used and the date has changed. This is similar to how __TIME__ is handled. If __DATE__ is present in the source code, ccache hashes the current date in order to be able to produce the correct object file if the __DATE__ macro affects the output. If you know that __DATE__ isn’t used in practise, or don’t care if ccache produces objects where __DATE__ is expanded to something in the past, you can set sloppiness to time_macros.

    • The __TIMESTAMP__ preprocessor macro is (potentially) being used and the source file’s modification time has changed. This is similar to how __TIME__ is handled. If __TIMESTAMP__ is present in the source code, ccache hashes the string representation of the source file’s modification time in order to be able to produce the correct object file if the __TIMESTAMP__ macro affects the output. If you know that __TIMESTAMP__ isn’t used in practise, or don’t care if ccache produces objects where __TIMESTAMP__ is expanded to something in the past, you can set sloppiness to time_macros.

    • The input file path has changed. Ccache includes the input file path in the direct mode hash to be able to take relative include files into account and to produce a correct object file if the source code includes a __FILE__ macro.

  • If a cache hit counter was not incremented even though the same code has been compiled and cached before, ccache has either detected that something has changed anyway or a cleanup has been performed (either explicitly or implicitly when a cache limit has been reached). Some perhaps unobvious things that may result in a cache miss are usage of __TIME__, __DATE__ or __TIMESTAMP__ macros, or use of automatically generated code that contains a timestamp, build counter or other volatile information.

  • If “Multiple source files” has been incremented, it’s an indication that the compiler has been invoked on several source code files at once. Ccache doesn’t support that. Compile the source code files separately if possible.

  • If “Unsupported compiler option” has been incremented, enable debug logging and check which compiler option was rejected.

  • If “Preprocessing failed” has been incremented, one possible reason is that precompiled headers are being used. See Precompiled headers for how to remedy this.

  • If “Could not use precompiled header” has been incremented, see Precompiled headers.

  • If “Could not use modules” has been incremented, see C++ modules.

Corrupt object files

It should be noted that ccache is susceptible to general storage problems. If a bad object file sneaks into the cache for some reason, it will of course stay bad. Some possible reasons for erroneous object files are bad hardware (disk drive, disk controller, memory, etc), buggy drivers or file systems, a bad prefix_command or compiler wrapper. If this happens, the easiest way of fixing it is this:

  1. Build so that the bad object file ends up in the build tree.

  2. Remove the bad object file from the build tree.

  3. Rebuild with CCACHE_RECACHE set.

An alternative is to clear the whole cache with ccache -C if you don’t mind losing other cached results.

There are no reported issues about ccache producing broken object files reproducibly. That doesn’t mean it can’t happen, so if you find a repeatable case, please report it.

More information

Credits, mailing list information, bug reporting instructions, source code, etc, can be found on ccache’s web site: https://ccache.dev.

Author

Ccache was originally written by Andrew Tridgell and is currently developed and maintained by Joel Rosdahl. See AUTHORS.txt or AUTHORS.html and https://ccache.dev/credits.html for a list of contributors.