GENINFO(1)

📦
version 2.6 · 15 versions
, in arch rolling

Okay bestie, this is geninfo(1) from the LCOV universe, and it's literally the manual page for geninfo. Buckle up, because this one is long and we are keeping every single detail. 💅

💖 NAME 💖

Yes, the page really has two NAME blocks, so here they both are, no cap:

  • geninfo — geninfo documentation
  • geninfo — Generate tracefiles from GCOV coverage data files

✨ SYNOPSIS ✨

geninfo
         [-h | --help]
         [--version]
         [-q | --quiet]
         [-v | --verbose]
         [--debug]
         [--comment *comment-string*]
         [-i | --initial]
         [--all]
         [-t | --test-name *test-name*]
         [-o | --output-filename *filename*]
         [-f | --follow]
         [-b | --base-directory *directory*]
         [--build-directory *directory*]
         [--branch-coverage]
         [--mcdc-coverage]
         [--checksum]
         [--no-checksum]
         [--compat-libtool]
         [--no-compat-libtool]
         [--gcov-tool *tool*]
         [--parallel | -j [*integer*]]
         [--large-file *regexp*]
         [--memory *integer_num_Mb*]
         [--msg-log [*log_file_name*]]
         [--ignore-errors *errors*]
         [--expect-message-count *message_type* = *expr* [,*message_type* = *expr* ..]]
         [--keep-going]
         [--preserve]
         [--filter *type*]
         [--demangle-cpp [param]]
         [--no-recursion]
         [--external]
         [--no-external]
         [--sort-input]
         [--config-file *config-file*]
         [--no-markers]
         [--profile [*profile-file*]]
         [--history-script *callback*]
         [--compat *mode* = on|off|auto]
         [--rc *keyword* = *value*]
         [--include *glob_pattern*]
         [--exclude *glob_pattern*]
         [--erase-functions *regexp_pattern*]
         [--substitute *regexp_pattern*]
         [--omit-lines *regexp_pattern*]
         [--fail-under-branches *percentage*]
         [--fail-under-lines *percentage*]
         [--forget-test-names]
         [--context-script *script_file*]
         [--criteria-script *script_file*]
         [--resolve-script *script_file*]
         [--version-script *script_file*]
         [--unreachable-script *module_file*]
         [--tempdir *dirname*]
         *directory*

🦋 DESCRIPTION 🦋

Use geninfo to create LCOV tracefiles from GCC and LLVM/Clang coverage data files. Heads up: see --gcov-tool for the considerations when you're working with LLVM. You can use genhtml to create an HTML report from a tracefile. 🎀

Note that geninfo is called by lcov --capture, so there is typically no need to call it directly. Like, it's giving "middle management energy" — it's doing important work behind the scenes. 💅

Unless the --output-filename option is specified, geninfo writes its output to one file with .info filename extension per input file.

Also note: the current user needs write access to both directory as well as to the original source code location. This is necessary because some temporary files have to be created there during the conversion process. So yeah, permissions check first, bestie. 🔑

By default, geninfo collects line and function coverage data. Neither branch nor MC/DC data is collected by default; you can use the --branch-coverage and --mcdc-coverage command line options to enable them, or you can permanently enable them by adding branch_coverage = 1 and/or mcdc_coverage = 1 to your personal, group, or site lcov configuration file. See lcovrc(5) for details. ✨

🌸 File types 🌸

A tracefile is a coverage data file in the format used by all LCOV tools such as geninfo, lcov, and genhtml. By convention, tracefiles have a .info filename extension. See “Tracefile format” below for a description of the file format. 🦋

A .gcda file is a compiler-specific file containing run-time coverage data. It is created and updated when a program compiled with GCC/LLVM’s --coverage option is run to completion. geninfo reads .gcda files in its default mode of operation. Note: earlier compiler versions used the .da filename extension for this file type. ⭐

A .gcno file is a compiler-specific file containing static, compile-time coverage data. It is created when source code is compiled with GCC/LLVM’s --coverage option. geninfo reads .gcno files when option --initial is specified. Note: earlier compiler versions used .bb and .bbg filename extensions for this file type. 🎀

A .gcov file is a textual or JSON representation of the data found in .gcda and .gcno files. It is created by the gcov tools that is part of GCC (see --gcov-tool for LLVM considerations). There are multiple gcov file format versions, including textual, intermediate, and JSON format. geninfo internally uses gcov to extract coverage data from .gcda and .gcno files using the best supported gcov file format. 👑

See the gcov man page for more information on .gcda, .gcno, and .gcov output formats.

🎀 Exclusion markers 🎀

To exclude specific lines of code from a tracefile, you can add exclusion markers to the source code. Similarly, you can mark specific regions of code as “unreachable”. An “unreachable” error message is generated if any coverpoints in unreachable regions are executed (i.e., have non-zero hit counts). See the retain_unreachable_coverpoints_if_executed section in lcovrc(1) for a description of the actions taken in this case. 🫶

Additionally, you can exclude specific branches or MC/DC expressions without excluding the involved lines from line and function coverage.

Exclusion markers are keywords which can for example be added in the form of a comment. See lcovrc(5) for how to override the exclusion keywords (e.g., to reuse markers inserted for other tools or to generate reports with different sets of excluded regions). 💜

The following markers are recognized by geninfo:

  • LCOV_EXCL_LINE — Lines containing this marker will be excluded.
  • LCOV_EXCL_START — Marks the beginning of an excluded section. The current line is part of this section.
  • LCOV_EXCL_STOP — Marks the end of an excluded section. The current line not part of this section.
  • LCOV_UNREACHABLE_LINE — If the marked line is ‘hit’, then generate an error: we believe the marked code is unreachable and so there is a bug in the code, the placement of the directive, or both. Lines containing this marker will be excluded from reporting. Apart from error reporting, this directive is equivalent to LCOV_EXCL_LINE.
  • LCOV_UNREACHABLE_START — Marks the beginning of an unreachable section of code. The current line in part of this region. As described in the LCOV_UNREACHABLE_LINE section, above: an error is generated if any code in the region is hit, but the code is excluded from reporting.
  • LCOV_UNREACHABLE_STOP — Marks the end of the region of unreachable code. The current line not part of this section.
  • LCOV_EXCL_BR_LINE — Lines containing this marker will be excluded from branch coverage.
  • LCOV_EXCL_BR_START — Marks the beginning of a section which is excluded from branch coverage. The current line is part of this section.
  • LCOV_EXCL_BR_STOP — Marks the end of a section which is excluded from branch coverage. The current line not part of this section.
  • LCOV_EXCL_EXCEPTION_BR_LINE — Lines containing this marker will be excluded from exception branch coverage: Exception branches will be ignored, but non-exception branches will not be affected.
  • LCOV_EXCL_EXCEPTION_BR_START — Marks the beginning of a section which is excluded from exception branch coverage. The current line is part of this section.
  • LCOV_EXCL_EXCEPTION_BR_STOP — Marks the end of a section which is excluded from exception branch coverage. The current line not part of this section.

In addition, geninfo also supports the --unreachable-script option. This provides a mechanism to exclude particular branch expressions and/or MC/DC conditions from coverage reports. A simple sample callback can be found in:

bin/unreach.pm

This sample callback adds support for branch-specific and condition-specific exclusion, specified by end-of-line annotations in the source code:

  • LCOV_EXCLUDE_BRANCH (expressionId[,blockId])+ — One or more space-separated branch exclusions may be specified.
    • expressionId: decimal integer which specifies which branch to exclude. An error is generated if the index is invalid.
    • blockId: decimal integer which specifies the block index which contains the excluded expression. An error is generated if the block index is invalid. If there is only one branch block at this location, then the block ID does not need to be specified.
  • LCOV_UNREACHABLE_COND ([groupSize,]conditionId(t|f))+ — One or more space-separated MC/DC exclusions may be specified.
    • groupSize: decimal integer which specifies the group which contains the excluded condition. An error is generated if the groupSize is invalid. If there is only one MC/DC group at this location, then the groupSize does not need to be specified.
    • conditionId: decimal integer which specifies which condition to exclude. An error is generated if the index is invalid.
    • sense: ‘t’ or ‘f’: the value of the condition which is unreachable.

For example, the line:

code // LCOV_EXCLUDE_BRANCH 0,1 3 LCOV_EXCLUDE_CONDITION 0f

will exclude branch expressions 1 and 3 from block 0 (or generate an error if there is more than one branch block on this line), and the ‘false’ sense of condition 0 (or generate an error if there is more than one MC/DC group on this line). 😭🔥

One way to determine the branch and/or condition ID in order to exclude them is to generate an HTML report without exclusions, then hover the mouse over the colorized branch/condition region and read the IDs from the tooltip popup. Another way is to simply read the BRDA/MCDC entries in coverage data file. 👑

See the comment at the top of the sample script for more details.

🔥 OPTIONS 🔥

In general, (almost) all geninfo options can also be specified in your personal, group, project, or site configuration file — see the --config-file section, below, and lcovrc(5) for details. ✨

  • -b, --base-directory directory — bestie, this tells geninfo to use directory as the base directory for relative paths. 💅

Use this one to specify the base directory of a build-environment when geninfo starts screaming error messages like:

ERROR: could not read source file /home/user/project/subdir1/subdir2/subdir1/subdir2/file.c

In that example you'd use /home/user/project as the base directory. Iconic, honestly.

This option is required when you're using geninfo on projects built with libtool or similar build environments that work with a base directory — meaning environments where the current working directory when invoking the compiler is not the same directory in which the source code file lives. 🫶

Heads up though: this option will not work in environments where multiple base directories are used. In that case use the configuration file setting geninfo_auto_base=1 (see lcovrc(5)). No cap.
  • --build-directory build_dir — search for .gcno data files from build_dir instead of finding them only adjacent to the corresponding .o and/or .gcda file. 🔍

By default, geninfo expects to find the .gcno and .gcda files (compile- and run-time data, respectively) in the same directory.

When this option is used:

geninfo path1 --build-directory path2 ...

then geninfo will look for the .gcno file

path2/relative/path/to/da_base.gcno

when it finds the .gcda file

path1/relative/path/to/da_base.gcda.

Use this option when you've used the GCOV_PREFIX environment variable to direct the gcc or llvm runtime environment to write coverage data files to somewhere other than the directory where the code was originally compiled. See gcc(1) and/or search for GCOV_PREFIX and GCOV_PREFIX_STRIP. ✨

This option can be used several times to specify multiple alternate directories to look for .gcno files. Super useful if your application uses code compiled in many separate locations — like common libraries shared between teams. 💖

  • --source-directory dirname — adds dirname to the list of places to look for source files. 🎀

For relative source file paths found in the gcov data — possibly after substitutions have been applied — geninfo will first look for the path from cwd (where genhtml was invoked) and then from each alternate directory name in the order specified. The first matching location is the one that gets used.

This option can be specified multiple times to add more directories to the source search path. Slay.

  • --branch-coverage — collect and retain branch coverage data. 🌿

This is equivalent to using the option --rc branch_coverage=1; the option was added to better match the genhtml interface.

  • --mcdc-coverage — collect and retain MC/DC data. 🔥

This is equivalent to using the option --rc mcdc_coverage=1. MC/DC coverage capture is supported for GCC versions 14.2 and higher, or LLVM versions 18.1 and higher.

See llvm2lcov --help for details on MC/DC data capture in LLVM.

See the MC/DC section of genhtml(1) for more details. 👑

  • --checksum, --no-checksum — specify whether to generate checksum data when writing tracefiles. 🧾

Use --checksum to enable checksum generation or --no-checksum to disable it. Checksum generation is disabled by default.

When checksum generation is enabled, a checksum will be generated for each source code line and stored along with the coverage data. This checksum will be used to prevent attempts to combine coverage data from different source code versions.

If you don't work with different source code versions, disable this option to speed up coverage data processing and to reduce the size of tracefiles. Speed = self care, honestly. 💅

Note that this option is somewhat subsumed by the --version-script option — which does something similar, but at the whole file level.

  • --compat mode = value [,mode = value,…] — set compatibility mode. 🛠️

Use --compat to specify that geninfo should enable one or more compatibility modes when capturing coverage data. You can provide a comma-separated list of mode=value pairs to specify the values for multiple modes.

Valid values are:

  • on — enable compatibility mode.
  • off — disable compatibility mode.
  • auto — apply auto-detection to determine if compatibility mode is required. Note that auto-detection is not available for all compatibility modes.

If no value is specified, on is assumed as the default value.

Valid modes are:

  • libtool — enable this mode if you're capturing coverage data for a project that was built using the libtool mechanism. See also --compat-libtool. The default value for this setting is on.
  • hammer — enable this mode if you're capturing coverage data for a project that was built using a version of GCC 3.3 that contains a modification (hammer patch) of later GCC versions. You can identify a modified GCC 3.3 by checking the build directory of your project for files ending in the extension .bbg. Unmodified versions of GCC 3.3 name these files .bb. The default value for this setting is auto.
  • split_crc — enable this mode if you're capturing coverage data for a project that was built using a version of GCC 4.6 that contains a modification (split function checksums) of later GCC versions. Typical error messages when running geninfo on coverage data produced by such GCC versions are out of memory and reached unexpected end of file. The default value for this setting is auto.
  • --compat-libtool, --no-compat-libtool — specify whether to enable libtool compatibility mode. 🧰

Use --compat-libtool to enable libtool compatibility mode or --no-compat-libtool to disable it. The libtool compatibility mode is enabled by default.

When libtool compatibility mode is enabled, geninfo will assume that the source code relating to a .gcda file located in a directory named .libs can be found in its parent directory.

If you have directories named .libs in your build environment but don't use libtool, disable this option to prevent problems when capturing coverage data. 🙈

  • --config-file config-file — specify a configuration file to use. See lcovrc(5) for details of the file format and options. Also see the config_file entry in the same man page for details on how to include one config file into another. 📄

When this option is specified, neither the system-wide configuration file /etc/lcovrc, nor the per-user configuration file ~/.lcovrc is read.

This option may be useful when there's a need to run several instances of geninfo with different configuration file options in parallel.

Note that this option must be specified in full — abbreviations are not supported. No shortcuts, bestie.
  • --profile [profile-data-file] — tell the tool to keep track of performance and other configuration data. 📊 If the optional profile-data-file is not specified, then the profile data is written to a file named with the same basename as the --output-filename, with suffix .json appended.

Profile data is useful if you're trying to optimize the geninfo implementation (see $LCOV_ROOT/share/lcov/support-scripts/spreadsheet.py), and can also enable faster geninfo --parallel execution (see the --history-script section of this man page).

  • --history-script callback — tell the tool to use performance data from a prior job to predict resource usage by the current job. 🔮 This may allow better segmentation to enable more balanced workloads between parallel threads — thus improving wall clock execution time.

A common source for the performance history is the previous-profile-data-file generated by the --profile argument.

See a sample callback implementation in $LCOV_ROOT/share/lcov/support-scripts/history.pm and its use in $LCOV_ROOT/share/lcov/example in the installed release (or .../example and .../scripts/history.pm in the source repository).

See genhtml(1) for more details. 🦋

  • --external, --no-external — specify whether to capture coverage data for external source files. 🌍

External source files are files which are not located in one of the directories specified by --directory or --base-directory. Use --external to include coverpoints in external source files while capturing coverage data or --no-external to exclude them. If your --directory or --base-directory path contains a soft link, then the actual target directory is not considered to be internal unless the --follow option is used.

The --no-external option is somewhat of a blunt instrument; the --exclude and --include options provide finer grained control over which coverage data is and is not included if your project structure is complex and/or --no-external doesn't do what you want.

Data for external source files is included by default. 💖

  • -f, --follow — follow links when searching .gcda files, as well as to decide whether a particular (symbolically linked) source directory is internal to the project or not — see the --no-external option above for more info. 🔗 The --follow command line option is equivalent to the geninfo_follow_symlinks config file option. See lcovrc(5) for more information.
  • --sort-input — bestie, this one controls whether file names get sorted before capture and/or aggregation. Why do we care? Because sorting reduces certain types of processing-order-dependent output differences ✨. See the sort_input section in lcovrc(5) <file://../man/lcovrc.5.html> for the deets.
  • --gcov-tool tool — tells geninfo where your gcov tool lives 🔍.

Okay so this one has lore. If you use --gcov-tool multiple times, the arguments get concatenated when the callback is executed — kind of like how the gcc -Xlinker parameter works. It's a possibly easier way to pass arguments to your tool without needing a whole wrapper script. In that case, your callback gets executed as: tool-0 ‘tool-1; … ‘filename’’. Note that the second and subsequent arguments are quoted when passed to the shell, so parameters containing spaces are handled 🫶.

The --gcov-tool argument may also be a split_char-separated string — see man(4) lcovrc.

A common use for this option is to enable LLVM:

geninfo --gcov-tool llvm-cov --gcov-tool gcov ...
geninfo --gcov-tool llvm-cov,gcov ...

Note: ‘llvm-cov gcov da_file_name’ will generate output in gcov-compatible format as required by lcov 💅.

If you don't specify it, ‘gcov’ is used by default.

  • -h, --help — prints a short help text, then dips out. That's it, that's the whole vibe 🎀.
  • --include pattern — include source files matching pattern.

Use this switch when you want coverage data for only a particular set of source files matching any of the given patterns. Multiple patterns? Just use multiple --include command line switches, easy. The patterns are interpreted as shell wildcard patterns (and heads up: you may need to escape them accordingly so your shell doesn't expand them first 🙈).

See the lcov man page for details.

  • --exclude pattern — exclude source files matching pattern.

Use this switch when you want to keep coverage data away from a particular set of source files matching any of the given patterns. Multiple patterns work the same way — multiple --exclude command line switches. Same deal with shell wildcard patterns (escape accordingly to stop the shell expanding them first). Note: the pattern must be specified to match the absolute path of each source file ⚠️.

Can be combined with the --include command line switch. But if a given file matches both the include pattern and the exclude pattern, the exclude pattern takes precedence — exclude always wins, periodt.

See the lcov man page for details.

  • --erase-functions regexp — exclude coverage data from lines which fall within a function whose name matches the supplied regexp. Note that this is a mangled or demangled name, depending on whether the –demangle-cpp option is used or not 👀.

Note that this option requires that you use a gcc version which is new enough to support function begin/end line reports, or that you configure the tool to derive the required data — see the derive_function_end_line discussion in the lcovrc man page.

  • --substitute regexp_pattern — applies the Perl regexp regexp_pattern to source file names found during processing. Super useful when the path name reported by gcov doesn't match your source layout and the file is not found 😭. See the lcov man page for more details.
  • --omit-lines regexp — exclude coverage data from lines whose content matches regexp.

Use this switch if you want to exclude line, branch, and MC/DC coverage data for some particular constructs in your code (e.g., some complicated macro). See the lcov man page for details.

  • --forget-test-names — if non-zero, ignore testcase names in the tracefile — i.e., treat all coverage data as if it came from the same testcase. This may improve performance and reduce memory consumption if you don't need per-testcase coverage summary in your coverage reports 🧠✨.

This option can also be configured permanently using the configuration file option forget_testcase_names.

  • --msg-log [log_file_name] — specify a location to store error and warning messages (in addition to writing to STDERR). If log_file_name is not specified, then the default location is used 📝.
  • --ignore-errors errors — specify a list of errors after which to continue processing.

Use this option to specify a list of one or more classes of errors after which geninfo should keep processing instead of aborting. Note that the tool will generate a warning (rather than a fatal error) unless you ignore the error two (or more) times:

geninfo ... --ignore-errors unused,unused

errors can be a comma-separated list of the following keywords:

  • branch — branch ID (2nd field in the .info file BRDA entry) does not follow expected integer sequence.
  • callback — version script error.
  • child — child process returned non-zero exit code during –parallel execution. This typically indicates that the child encountered an error: see the log file immediately above this message. In contrast: the parallel error indicates an unexpected/unhandled exception in the child process — not a typical lcov error.
  • corrupt — corrupt/unreadable file found.
  • count — an excessive number of messages of some class have been reported — subsequent messages of that type will be suppressed. The limit can be controlled by the ‘max_message_count’ variable. See the lcovrc man page.
  • deprecated — you are using a deprecated option. This option will be removed in an upcoming release — so you should change your scripts now ⚠️.
  • empty — the .info data file is empty (e.g., because all the code was ‘removed’ or excluded.
  • excessive — your coverage data contains a suspiciously large ‘hit’ count which is unlikely to be correct — possibly indicating a bug in your toolchain. See the excessive_count_threshold section in lcovrc(5) <file://../man/lcovrc.5.html> for details.
  • fork — unable to create child process during –parallel execution. If the message is ignored (–ignore-errors fork), then genhtml will wait a brief period and then retry the failed execution. If you see continued errors, either turn off or reduce parallelism, set a memory limit, or find a larger server to run the task 😮‍💨.
  • format — unexpected syntax or value found in .info file — for example, negative number or zero line number encountered.

A count field which is not a legal integer (for example ‘nan’ or ‘inf’) is likewise treated as zero if this message is ignored; a count which is a number but not an integer is truncated toward zero, and one too large to store is clamped at 9223372036854775807. Numbers, babes 🔢.

  • gcov — the gcov tool returned with a non-zero return code.
  • graph — the graph file could not be found or is corrupted.
  • inconsistent — your coverage data is internally inconsistent: it makes two or more mutually exclusive claims. For example, some expression is marked as both an exception branch and not an exception branch. (See genhtml(1) <file://../man/genhtml.1.html> for more details.
  • internal — internal tool issue detected. Please report this bug along with a testcase 🙏.
  • mismatch — incorrect information found in coverage data and/or source code — for example, the source code contains overlapping exclusion directives.
  • missing — file does not exist or is not readable.
  • negative — negative ‘hit’ count found.

A negative count is not a legal value for any cover type, so if this message is ignored, the count is treated as zero.

Note that negative counts may be caused by a known GCC bug — see

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68080

and try compiling with “-fprofile-update=atomic”. You will need to recompile, re-run your tests, and re-capture coverage data 💀.

  • package — a required perl package is not installed on your system. In some cases, it is possible to ignore this message and continue — however, certain features will be disabled in that case.
  • parallel — various types of errors related to parallelism — i.e., a child process died due to an error. The corresponding error message appears in the log file immediately before the parallel error.

If you see an error related to parallel execution that seems invalid, it may be a good idea to remove the –parallel flag and try again. If removing the flag leads to a different result, please report the issue (along with a testcase) so that the tool can be fixed 🛠️.

  • parent — the parent process exited while child was active during –parallel execution. This happens when the parent has encountered a fatal error — e.g. an error in some other child which was not ignored. This child cannot continue working without its parent — and so will exit 😔.
  • path — some file paths were not resolved — e.g., .gcno file corresponding to some .gcda was not found; see –build-directory option for additional information.
  • range — coverage data refers to a line number which is larger than the number of lines in the source file. This can be caused by a version mismatch or by an issue in the gcov data.
  • source — the source code file for a data set could not be found.
  • unreachable — a coverpoint (line, branch, function, or MC/DC) that lives inside an “unreachable” region actually got executed (hit). Which means either the code is wrong, the directive placement is wrong, or honestly both 🙈. If you tell it to ignore this error, then whether the offending coverpoint gets retained (not excluded) or not depends on the value of the retain_unreachable_coverpoints_if_executed configuration parameter. See lcovrc(5) and the “Exclusion markers” section up above, bestie.
  • unsupported — the feature you're asking for just isn't supported for this tool configuration 😭. Example: function begin/end line range exclusions use some GCOV features that older GCC releases simply do not have. No cap.
  • unused — the include/exclude/erase/omit/substitute pattern didn't match any file pathnames. It's giving "we dressed up but nobody came" 💀.
  • usage — unsupported usage detected — like, an unsupported option combination. You combined things that do not combine, queen.
  • utility — a tool called during processing returned an error code (e.g., find encountered an unreadable directory).
  • version — revision control IDs of the file which we are trying to merge are not the same — so line numbering and other information may be incorrect. Big yikes.

Also see the --ignore-errors section in genhtml(1). The description over there may be more complete and/or more fully explained. Go read it, it's a whole thing 🫶.

See lcovrc(5) for a discussion of the max_message_count parameter which can be used to control the number of warnings which are emitted before all subsequent messages are suppressed. This can be used to reduce log file volume — very much a "quiet the group chat" move 🔇.

  • --expect-message-count message_type=expr [,message_type=expr]— givegeninfoa constraint on the number of messages of one or more types which are expected to be produced during execution. If the constraint is not true, then generate an error of typecount` (see above). It's literally a vibe check for your messages ✨.

See genhtml(1) for more details about the flag, as well as the expect_message_count section in lcovrc(5) for a description of the equivalent configuration file option.

  • --keep-going — do not stop if error occurs: attempt to generate a result, however flawed 💅. It's giving "we ball, even if we ball badly."

This command line option corresponds to the stop_on_error [0|1] lcovrc option. See lcovrc(5) for more details.

  • --fail-under-lines percentage — use this option to tell geninfo to exit with a status of 1 if the total line coverage is less than percentage. See lcov(1) for more details. Basically a "you must be at least this tall to ride" gate 🎢.
  • --preserve — preserve intermediate data files (e.g., for debugging). By default, intermediate files are deleted. So without this, they're gone, poof, ghosted 🦋.
  • --filter filters — specify a list of coverpoint filters to apply to input data. See the genhtml man page for details.
  • --demangle-cpp [param] — demangle C++ method and function names in captured output. See the genhtml man page for details.
  • -i, --initial — capture initial zero coverage data. Run geninfo with this option on the directories containing .bb, .bbg or .gcno files before running any test case. The result is a “baseline” coverage data file that contains zero coverage for every instrumented line and function. Combine this data file (using lcov -a) with coverage data files captured after a test run to ensure that the percentage of total lines covered is correct even when not all object code files were loaded during the test. Also see the --all flag, below 👇.
Note: the --initial option is not supported for gcc versions less than 6, and does not generate branch coverage information for gcc versions less than 8. Old gcc? Old news, no baseline for you 😭.
  • --all — capture coverage data from both compile time (.gcno) data files which do not have corresponding runtime (.gcda) data files, as well as from those that do have corresponding runtime data. There will be no runtime data unless some executable which links the corresponding object file has run to completion.

Note that the execution count of coverpoints found only in files which do not have any runtime data will be zero.

This flag is ignored if the --initial flag is set.

Using the --all flag is equivalent to executing both geninfo --initial … and geninfo … and merging the result.

Also see the geninfo_capture_all entry in man(5) lcovrc. It's literally the "get everybody, even the ones who didn't show up" flag 👑.

  • --no-markers — unless the --no-markers option is used, geninfo will apply both region and branch_region filters to the captured coverage data. Use this option if you want to get coverage data without regard to exclusion markers in the source code file.

If any --filter options are applied, then the default region filters are not used.

--no-markers should not be specified along with --filter. Do not do it. I'm serious. 🙈
  • --no-recursion — use this option if you want to get coverage data for the specified directory only without processing subdirectories. Just the one level, no rabbit hole 🐇.
  • -o, --output-filename output-filename — write all data to output-filename. If you want to have all data written to a single file (for easier handling), use this option to specify the respective filename. By default, one tracefile will be created for each processed .gcda file. One file to rule them all vs. a whole pile — your call 💅.
  • --context-script script — use script to collect additional tool execution context information — to aid in infrastructure debugging and/or tracking.

See the genhtml man page for more details on the context script.

  • --criteria-script script — use script to test for coverage acceptance criteria.

See the genhtml man page for more details on the criteria script. Note that geninfo does not keep track of date and owner information (see the --annotate-script entry in the genhtml man page) — so this information is not passed to the geninfo callback. F in the chat for the missing metadata 😭.

  • --resolve-script script — use script to find the file path for some source or GCNO file which appears in an input data file if the file is not found after applying --substitute patterns and searching the --source-directory or --build-directory list. It's the "have you tried asking a script where your file is" option 🕵️‍♀️.

This option is equivalent to the resolve_script config file option. In addition, the geninfo_follow_path_links config file option can be used to resolve source paths to their actual target.

See lcovrc(5) for details.

  • --version-script script — use script to get a source file's version ID from revision control when extracting data. The ID is used for error checking when merging .info files.

See the genhtml man page for more details on the version script.

  • --unreachable-script module — use module to decide whether particular branch expressions and/or MC/DC conditions should be removed from the coverage report. This option is equivalent to the unreachable_script config file option. See lcovrc(5) for details.
Note that “module” is required to be a Perl module 🧚 — not a script, not vibes, a module.

See the genhtml man page and the “Exclusion markers” section, above, for more information.

  • -v, --verbose — increment informational message verbosity. This is mainly used for script and/or flow debugging — e.g., to figure out which data file are found, where. Also see the --quiet flag. It's the "tell me everything" mode, main character energy 👑

Messages are sent to stdout unless there is no output file (i.e., if the coverage data is written to stdout rather than to a file) and to stderr otherwise. Little routing detail, but lowkey important.

  • -q, --quiet — decrement informational message verbosity. Decreased verbosity will suppress ‘progress’ messages for example — while error and warning messages will continue to be printed. So it gets quieter, but the important mess stays loud 🔥.
  • --debug — increment ‘debug messages’ verbosity. This is useful primarily to developers who want to enhance the lcov tool suite. Devs only, bestie 🫶.
  • --comment comment_string — append comment_string to list of comments emitted into output result file. This option may be specified multiple times. Comments are printed at the top of the file, in the order they were specified.

Comments can be useful to document the conditions under which the trace file was generated: host, date, environment, etc. Basically a little diary entry at the top of your file 📝💖.

  • --parallel [integer], -j [integer] — specify parallelism to use during processing (maximum number of forked child processes). If the optional integer parallelism parameter is zero or is missing, then use to use up the number of cores on the machine. Default is to use a single process (no parallelism). Haters will say it's slow. It's not 💅

The --large-file option described below may be necessary to enable parallelism to succeed in the presence of data files which consume excessive memory in gcov.

Note that geninfo writes coverage data (.info) files rather than reading them, so the parallel .info file read controlled by the parallel_parse_min_lines and parallel_parse_chunks_per_worker entries in lcovrc(5) does not apply here; see the --parallel sections of lcov(1) and genhtml(1). Yeah, ikyk 😌

Also see the memory, memory_percentage, max_fork_fails, fork_fail_timeout, geninfo_chunk_size, geninfo_dedicate_segment_size, dedicate_segment_threshold, dedicate_segment_line_estimate and geninfo_interval_update entries in lcovrc(5) for the tea on some options that can help with parameter tuning and performance optimization. Specifically, geninfo_dedicate_segment_size (and, when --history-script data is available, dedicate_segment_threshold) arranges for a large compilation unit to get its own child process and be scheduled first, so it doesn't serialize the tail of the run. A previously generated execution profile can help enable better utilization and faster parallel execution — see the --profile and --history-script sections of this man page. 💅

  • --large-file regexp — GCDA files whose name matches a --large-file regexp are processed serially — not in parallel with other files — so that their gcov process can use all available system memory. 🫶

Use this option if you see errors related to memory allocation from gcov. (real "oh no" energy, bestie)

This feature is exactly as if you had moved the matching GCDA files to another location and processed them serially, then processed the remaining GDCA files in parallel and merged the results. Yes, GDCA. That's what it says. Copying it exactly. 🙈

Note that this is a memory-safety control, distinct from the latency-oriented geninfo_dedicate_segment_size: a --large-file match runs serially in the parent process, whereas a file that only exceeds geninfo_dedicate_segment_size is still forked (in its own dedicated child) and processed in parallel. A file matched by --large-file is never also given a dedicated child.

This option may be used multiple times to specify more than one regexp. ✨

  • --memory integer — specify the maximum amount of memory to use during parallel processing, in Mb. Effectively, the process will not fork() if this limit would be exceeded. Default is 0 (zero) — which means there is no limit. 😌

This option may be useful if the compute farm environment imposes strict limits on resource utilization such that the job will be killed if it tries to use too many parallel children — but the user does not know a priori what the permissible maximum is. This option enables the tool to use maximum parallelism, up to the limit imposed by the memory restriction. Big "work smarter not harder" energy.

The configuration file memory_percentage option provides another way to set the maximum memory consumption. See lcovrc(5) for details.

  • --rc keyword = value — override a configuration directive. 🎀

Use this option to specify a keyword = value statement which overrides the corresponding configuration statement in the lcovrc configuration file. You can specify this option more than once to override multiple configuration statements. See lcovrc(5) for a list of available keywords and their meaning.

  • -t, --test-name testname — use test case name testname for the resulting data. Valid test case names can consist of letters, decimal digits and the underscore character (_). 👑

This proves useful when data from several test cases is merged (i.e. by simply concatenating the respective tracefiles), in which case a test name can be used to differentiate between data from each test case. Lowkey such a glow-up for your tracefiles.

  • --version — print version number, then exit. That's it. That's the whole vibe check. ⭐
  • --tempdir dirname — write temporary and intermediate data under the indicated directory. Default is the platform temporary directory: $TMPDIR, $TEMP or $TMP if one of them names a writeable directory, otherwise /tmp on Unix or C:\temp on Windows. 🦋

A uniquely named subdirectory of dirname is created for the run, and the data is written there — so several runs can share one dirname without colliding. dirname itself is created if it does not exist.

Unless --preserve is specified, the generated subdirectory and its contents are removed when geninfo exits. 🧹

This option can also be configured permanently using the configuration file option lcov_tmp_dir.

🌈 TRACEFILE FORMAT 🌈

Following is a quick description of the tracefile format as used by genhtml, geninfo and lcov.

A tracefile is made up of several human-readable lines of text, divided into sections. Human-readable!! Take notes, every other binary format ever.

Where a record appears is part of the format, and not merely a convention:

   tracefile := ( comment | blank )*  section*
   section   := 'TN:'?  'SF:'  ( coverpoint | 'VER:' | comment | blank )*
                'end_of_record'

That is: 💖

  • A section begins with an SF: record (KF: is accepted as a synonym) and ends with an end_of_record record.
  • An optional TN: record precedes the SF: record of the section whose data it names. TN: identifies the testcase which the section's data belongs to, so it cannot appear after the data it applies to. (order matters, bestie — no take-backs)
  • The VER: record and all coverpoint records (FNL:, FNA:, FNDA:, BRDA:, MCDC:, DA: and the corresponding *F:/*H: summary records) appear within a section, in any order. They mean nothing outside one.
  • A comment record or a blank line may appear anywhere — inside a section or outside one. 🍓

Anything else — a coverpoint record outside any section, a TN: record inside one, a second SF: with no intervening end_of_record, an end_of_record with no section open, or end of file with a section still open — is reported as an ERROR_FORMAT. See the --ignore-errors format discussion in lcov(1). If that error is ignored, the reader recovers by discarding the misplaced record (or, in the unterminated-section cases, by closing the section that is open) and keeping the data which was in the right place. Graceful recovery. We love a bounce-back.

Records of a cover type which is turned off are discarded as soon as they are recognized, and their contents are not examined: with branch coverage disabled, for example, a malformed BRDA: record is dropped in silence rather than reported as an ERROR_FORMAT. The check is what costs the time, and the data would be thrown away in either case. So whether a malformed coverpoint record is diagnosed depends on whether its cover type is enabled — see branch_coverage, mcdc_coverage and function_coverage in lcovrc(5). Where a record appears is still checked in every case: a BRDA: record outside any section is an ERROR_FORMAT whether or not branch coverage is enabled. 💀

A comment record is:

#comment_string

There is no space before or after the # character. Comment records are ignored when a tracefile is read: they are not coverage data, so nothing is retained and nothing is written back out. If the --comment comment_string option is supplied, then a comment record appears at the top of the tracefile which the tool writes. ✨

If available, a tracefile begins with the testname, which is stored in the following format:

TN:<test name>

For each source file referenced in the .gcda file, there is a section containing filename and coverage data:

SF:<path to the source file>

An optional source code version ID follows:

VER:<version ID>

If present, the version ID is compared before file entries are merged (see lcov --add-tracefile), and before the 'source detail' view is generated by genhtml. See the --version-script callback_script documentation and the sample usage in the lcov regression test examples. 🎀

Function coverage data follows. Note that the format of the function coverage data has changed from LCOV 2.2 onward. The tool continues to be able to read the old format but now writes only the new format. This change was made so that function filter outcome is persistent in the generated tracefile. (iconic behaviour, honestly)

Functions and their aliases are recorded contiguously:

First, the leader:

FNL:<index>,<line number of function start>[,line number of function end>]

Then the aliases of the function; there will be at least one alias. All aliases of a particular function share the same index.

FNA:<index>,<execution count>,<name>

The now-obsolete function data format is:

FN:<line number of function start>,[<line number of function end>,]<function name>

The 'end' line number is optional, and is generated only if the compiler/toolchain version is recent enough to generate the data (e.g., gcc 9 or newer). This data is used to support the --erase-functions and --show-proportions options. If the function end line data is not available, then these features will not work. No end line, no features. That's the deal.

Next, there is a list of execution counts for each instrumented function:

FNDA:<execution count>,<function name>

This list is followed by two lines containing the number of functions found and hit:

FNF:<number of functions found>
FNH:<number of function hit>

Note that, as of LCOV 2.2, these numbers count function groups — not the individual aliases. 👑

Branch coverage information is stored one line per branch:

BRDA:<line_number>,[<exception>][<fallthrough][<unreachable>]<block>,<branch>,<taken>

(yes, we're copying that bracket situation exactly as written — not our business, not our problem 💅)

<line_number>

is the line number where the branch is found — and is expected to be a non-zero integer.

<block> and <branch>

serve to uniquely define a particular edge in the expression tree of a particular conditional found on the associated line.

Within a particular line, <block> is an integer numbered from zero with no gaps. For some languages and some coding styles, there will only be one block (index value zero) on any particular line. 🦋

<branch> is a string which serves to uniquely identify a particular edge. For some languages and tools — e.g., C/C++ code compiled with gcc or llvm — <branch> is an ordered integer index related to expression tree traversal order of the associated conditional. For others, it may be a meaningful string — see below. <branch> appears in the 'tooltip' popup of the associated branch in the genhtml output — so human-readable values are helpful to users who are trying to understand coverage results — for example, in order to develop additional regression tests, to improve coverage. Be nice to your future self and name things well ✨

  • <taken> — either - if the corresponding expression was never evaluated (like, the basic block containing the branch never ran at all 😴) or a number telling you how often that branch was taken.
  • <exception> — it's e (yes, just one character) if this is a branch related to exception handling, and it's simply not there if the branch has nothing to do with exceptions. Heads up tho: identifying exception branches requires compiler support, so geninfo can only spot them if your toolchain version is new enough to actually support the feature. iykyk.
  • <fallthrough> — it's f (again, one character) if this branch is marked as ‘fallthrough’ (like in the gcov output), and it's absent if the branch isn't marked. Same deal as above: fallthrough branch identification requires compiler support, and geninfo will only pick them up if your toolchain version is new enough to support the feature. Also, a branch can be marked as ‘<exception>’ or ‘<fallthough’ — but not both, bestie, pick a lane 💅.
  • <unreachable> — it's U (single character) if this branch is associated with an ‘unreachable directive, and it just isn't present if the branch is reachable. The whole point of unreachability is to mark branch expressions that should not show up in coverage reports — so you can ignore uncovered branches that could never be covered anyway and actually focus on the parts of the report that matter. 🌸

Peep the “–unreachable-script” discussion in genhtml(1) <file://../man/genhtml.1.html> and the “ignore_unreachable_flag” discussion in lcovrc(5) <file://../man/lcovrc.5.html> for the full tea on how branch exclusion markers get set and used.

So these next ones are example branch records where the <branch> expression values are human-readable strings:

BRDA:10,0,enable,1
BRDA:10,0,!enable,0

In this case the corresponding code from line 10 is very likely similar to:

if (enable) {
  ...
}

such that the associated testcase entered the block (‘enable’ evaluated to ‘true’). Slay. ✨

Arbitrarily complicated branch expressions are supported — including branch expressions that contain commas (like, say, an expression with a function call in it). No cap.

⚠️ Note that particular tools may or may not suppress expressions which are statically true or statically false — e.g., expressions using template parameters. Which makes it potentially complicated to compare coverage data generated by two different tools. Fr fr, watch out for that one. 😭

a note on branch identification, matching, and merging 🫶

As we said before: lcov merely presents coverage data that gets identified by other tools such as gcov. Unfortunately, gcov (and other tools) do not produce sufficient information to uniquely identify branch expressions — so lcov is forced to use a heuristic solution based on the order of appearance plus the number and type of individual branch elements — say, in the gcov output.

lcov computes a ‘signature’ for each “<block>”. If the block ID and signature found while capturing data from multiple files (or while merging coverage data) match — then the block is considered identical and the corresponding expression counts get combined. 💖

Within a particular line, there may be one or more blocks with the same signature — for example, corresponding to independent branch expressions in function call arguments on that line, or to different macro or template expansion. lcov keeps track of the order that branch blocks were found (say, in the gcov output or in an lcov .info file).

When merging branch data on a particular line:

  • Blocks which have different signatures are considered to be distinct.
  • Blocks which have identical signatures are matched in order: the zeroth block with signature s1 in coverage DB A is merged with the zeroth block with signature s1 in coverage DB B — and so on.
  • If one or the other DB does not contain some signature, or one DB contains more blocks with the same signature: then those non-matching blocks are simply copied to the merged output.
💀 This heuristic can cause ambiguity in the generated report — especially for cases where a particular line expands in different ways in different compilation units (e.g., due to template or macro expansion). In such cases it may just not be possible for lcov to know that some block whose signature and order appear to match is not actually the same block. Lowkey you can sometimes refactor the code slightly to introduce explicit temporaries (on their own line) to resolve the ambiguity. 🧚

Branch coverage summaries are stored in two lines:

BRF:<number of branches found>

Note that this count does not include ‘excluded’ branches. (Excluded branches do still appear as BRDA records, tho. 🙈)

BRH:<number of branches hit>

Note that this count does not include ‘excluded’ branches. Excluded branches do appear as BRDA records.

MC/DC information is stored one line per expression:

MCDC:<line_number>,[<unreachable>]<groupSize>,<sense>,<taken>,<index>,<expression>

where:

  • <line_number> — is the line number where the condition is found — and is expected to be a non-zero integer.
  • <groupSize> and <index> — these two serve to uniquely define a particular element in the expression tree of a particular conditional found on the associated line.

Within a particular line and group:

  • <index> — is an integer numbered from zero to <groupSize> - 1 with no gaps. For some languages and some coding styles, there will only be one group on any particular line.
  • <sense> — is either “f” or “t”, indicating whether the condition is sensitive to the indicated change — that is, does the condition outcome change if the corresponding changes from ‘false’ to ‘true’ or from ‘true to false, respectively.
  • <taken> — is a count: 0 (zero) if the expression was not sensitized, non-zero if it was sensitized. Note that some tools may treat <taken> as the number of times the expression was sensitized where others may treat it as a boolean — 1:sensitized or 0: not sensitized.
  • <expression> — is an arbitrary string, intended to be a meaningful string which will help the user understand the condition context — see below. <expression> appears in the ‘tooltip’ popup of the associated MC/DC condition in the genhtml output — so human-readable values are genuinely helpful to users trying to understand coverage results — for example, in order to develop additional regression tests, to improve coverage. 👑

For a given <groupSize> and <index>, the <expression> should be identical for both “t” and “f” senses.

  • <unreachable> — is U (single character) if this condition is associated with an ‘unreachable directive, and it's not present if the condition is reachable. Unreachability is intended to mark conditions which should not appear in coverage reports — enabling users to ignore uncovered conditions which can never be covered, so they can concentrate on other areas of the coverage report.

See the “–unreachable-script” discussion in genhtml(1) <file://../man/genhtml.1.html> and the “ignore_unreachable_flag” discussion in lcovrc(5) <file://../man/lcovrc.5.html> for more information on how branch exclusion markers are set and used.

The following are example MC/DC records whose <expression> values are human-readable strings:

MCDC:10,2,f,0,0,enable
MCDC:10,2,t,1,0,enable
...

In this case, the corresponding code from line 10 is very likely similar to:

if (enable ...) {
  ...
}

such that the associated testcase was sensitive to a change of ‘enable’ from true to false (but not the converse). Iconic behavior honestly. 🦋

Arbitrarily complicated expressions are supported — including expressions which contain commas (e.g., in an expression containing a function call).

⚠️ Note that particular tools may or may not suppress expressions which are statically true or statically false — e.g., expressions using template parameters. This makes it potentially complicated to compare coverage data generated by two different tools. Yeah, it's giving "good luck comparing" 😭

MCDC coverage summaries are stored in two lines:

MCF:<number of conditions found>

Note that this count does not include ‘excluded’ conditions. Excluded branches do appear as MCDC records.

MCH:<number of condition hit>

Note that this count does not include ‘excluded’ conditions. Excluded branches do appear as MCDC records.

Then there's a list of execution counts for each instrumented line (i.e. a line which resulted in executable code):

DA:<line number>,<execution count>[,<checksum>]

Note that there may be an optional checksum present for each instrumented line. The current geninfo implementation uses an MD5 hash as checksumming algorithm. 🔥

At the end of a section, there's a summary about how many lines were found and how many were actually instrumented:

LH:<number of lines with a non-zero execution count>
LF:<number of instrumented lines>

Each section ends with:

end_of_record

In addition to the main source code file, there are sections for all #included files which also contain executable code. Periodt. 💅

📌 Note that the absolute path of a source file is generated by interpreting the contents of the respective .gcno file (see gcov(1) <file://../man/gcov.1.html> for more information on this file type). Relative filenames are prefixed with the directory in which the .gcno file is found.
📌 Note also that symbolic links to the .gcno file will be resolved so that the actual file path is used instead of the path to a link. This approach is necessary for the mechanism to work with the /proc/gcov files. Lowkey crucial, ngl.

🌈 FILES 🌈

  • /etc/lcovrc — the system-wide configuration file. 🎀
  • ~/.lcovrc — the per-user configuration file. 💖
  • bin/getp4version — sample script for use with --version-script that obtains version IDs via Perforce.
  • bin/get_signature — sample script for use with --version-script that uses md5hash as version IDs. ✨

👑 AUTHOR 👑

  • Peter Oberparleiter <<Peter.Oberparleiter@de.ibm.com>>
  • Henry Cox <<henry.cox@mediatek.com>>
    • Filtering, error management, parallel execution sections. 🫶

✨ SEE ALSO ✨

okay so when you're done here and you're like "wait, what about the rest of the toolchain??" — bestie, there's a whole squad 💅

  • genhtml(1) — gotta turn that data into pretty HTML? she does that
  • lcov(1) — the main character of the whole operation fr fr
  • lcovrc(5) — all the config file lore, iykyk
  • gcov(1) — the OG, the blueprint, the source of the numbers
  • llvm2lcov(1) — for the LLVM girlies 🦋

and if you wanna vibe with the repo itself:

https://github.com/linux-test-project/lcov

like, go star it or whatever ⭐

2024-2026, LCOV Project — protect the project besties 😭👑

this page is from lcov version 2.6, dated 2026-10-03 — that's the GENINFO(1) manual, and that's the whole thing, no cap 💖✨🫶
📦 where this page comes from