test-lib: add a GIT_TEST_PASSING_SANITIZE_LEAK=check mode

[git/debian.git] / t / README

Core Git Tests
==============

This directory holds many test scripts for core Git tools.  The
first part of this short document describes how to run the tests
and read their output.

When fixing the tools or adding enhancements, you are strongly
encouraged to add tests in this directory to cover what you are
trying to fix or enhance.  The later part of this short document
describes how your test scripts should be organized.


Running Tests
-------------

The easiest way to run tests is to say "make".  This runs all
the tests.

    *** t0000-basic.sh ***
    ok 1 - .git/objects should be empty after git init in an empty repo.
    ok 2 - .git/objects should have 3 subdirectories.
    ok 3 - success is reported like this
    ...
    ok 43 - very long name in the index handled sanely
    # fixed 1 known breakage(s)
    # still have 1 known breakage(s)
    # passed all remaining 42 test(s)
    1..43
    *** t0001-init.sh ***
    ok 1 - plain
    ok 2 - plain with GIT_WORK_TREE
    ok 3 - plain bare

Since the tests all output TAP (see http://testanything.org) they can
be run with any TAP harness. Here's an example of parallel testing
powered by a recent version of prove(1):

    $ prove --timer --jobs 15 ./t[0-9]*.sh
    [19:17:33] ./t0005-signals.sh ................................... ok       36 ms
    [19:17:33] ./t0022-crlf-rename.sh ............................... ok       69 ms
    [19:17:33] ./t0024-crlf-archive.sh .............................. ok      154 ms
    [19:17:33] ./t0004-unwritable.sh ................................ ok      289 ms
    [19:17:33] ./t0002-gitfile.sh ................................... ok      480 ms
    ===(     102;0  25/?  6/?  5/?  16/?  1/?  4/?  2/?  1/?  3/?  1... )===

prove and other harnesses come with a lot of useful options. The
--state option in particular is very useful:

    # Repeat until no more failures
    $ prove -j 15 --state=failed,save ./t[0-9]*.sh

You can give DEFAULT_TEST_TARGET=prove on the make command (or define it
in config.mak) to cause "make test" to run tests under prove.
GIT_PROVE_OPTS can be used to pass additional options, e.g.

    $ make DEFAULT_TEST_TARGET=prove GIT_PROVE_OPTS='--timer --jobs 16' test

You can also run each test individually from command line, like this:

    $ sh ./t3010-ls-files-killed-modified.sh
    ok 1 - git update-index --add to add various paths.
    ok 2 - git ls-files -k to show killed files.
    ok 3 - validate git ls-files -k output.
    ok 4 - git ls-files -m to show modified files.
    ok 5 - validate git ls-files -m output.
    # passed all 5 test(s)
    1..5

You can pass --verbose (or -v), --debug (or -d), and --immediate
(or -i) command line argument to the test, or by setting GIT_TEST_OPTS
appropriately before running "make". Short options can be bundled, i.e.
'-d -v' is the same as '-dv'.

-v::
--verbose::
        This makes the test more verbose.  Specifically, the
        command being run and their output if any are also
        output.

--verbose-only=<pattern>::
        Like --verbose, but the effect is limited to tests with
        numbers matching <pattern>.  The number matched against is
        simply the running count of the test within the file.

-x::
        Turn on shell tracing (i.e., `set -x`) during the tests
        themselves. Implies `--verbose`.
        Ignored in test scripts that set the variable 'test_untraceable'
        to a non-empty value, unless it's run with a Bash version
        supporting BASH_XTRACEFD, i.e. v4.1 or later.

-d::
--debug::
        This may help the person who is developing a new test.
        It causes the command defined with test_debug to run.
        The "trash" directory (used to store all temporary data
        during testing) is not deleted even if there are no
        failed tests so that you can inspect its contents after
        the test finished.

-i::
--immediate::
        This causes the test to immediately exit upon the first
        failed test. Cleanup commands requested with
        test_when_finished are not executed if the test failed,
        in order to keep the state for inspection by the tester
        to diagnose the bug.

-l::
--long-tests::
        This causes additional long-running tests to be run (where
        available), for more exhaustive testing.

-r::
--run=<test-selector>::
        Run only the subset of tests indicated by
        <test-selector>.  See section "Skipping Tests" below for
        <test-selector> syntax.

--valgrind=<tool>::
        Execute all Git binaries under valgrind tool <tool> and exit
        with status 126 on errors (just like regular tests, this will
        only stop the test script when running under -i).

        Since it makes no sense to run the tests with --valgrind and
        not see any output, this option implies --verbose.  For
        convenience, it also implies --tee.

        <tool> defaults to 'memcheck', just like valgrind itself.
        Other particularly useful choices include 'helgrind' and
        'drd', but you may use any tool recognized by your valgrind
        installation.

        As a special case, <tool> can be 'memcheck-fast', which uses
        memcheck but disables --track-origins.  Use this if you are
        running tests in bulk, to see if there are _any_ memory
        issues.

        Note that memcheck is run with the option --leak-check=no,
        as the git process is short-lived and some errors are not
        interesting. In order to run a single command under the same
        conditions manually, you should set GIT_VALGRIND to point to
        the 't/valgrind/' directory and use the commands under
        't/valgrind/bin/'.

--valgrind-only=<pattern>::
        Like --valgrind, but the effect is limited to tests with
        numbers matching <pattern>.  The number matched against is
        simply the running count of the test within the file.

--tee::
        In addition to printing the test output to the terminal,
        write it to files named 't/test-results/$TEST_NAME.out'.
        As the names depend on the tests' file names, it is safe to
        run the tests with this option in parallel.

-V::
--verbose-log::
        Write verbose output to the same logfile as `--tee`, but do
        _not_ write it to stdout. Unlike `--tee --verbose`, this option
        is safe to use when stdout is being consumed by a TAP parser
        like `prove`. Implies `--tee` and `--verbose`.

--with-dashes::
        By default tests are run without dashed forms of
        commands (like git-commit) in the PATH (it only uses
        wrappers from ../bin-wrappers).  Use this option to include
        the build directory (..) in the PATH, which contains all
        the dashed forms of commands.  This option is currently
        implied by other options like --valgrind and
        GIT_TEST_INSTALLED.

--no-bin-wrappers::
        By default, the test suite uses the wrappers in
        `../bin-wrappers/` to execute `git` and friends. With this option,
        `../git` and friends are run directly. This is not recommended
        in general, as the wrappers contain safeguards to ensure that no
        files from an installed Git are used, but can speed up test runs
        especially on platforms where running shell scripts is expensive
        (most notably, Windows).

--root=<directory>::
        Create "trash" directories used to store all temporary data during
        testing under <directory>, instead of the t/ directory.
        Using this option with a RAM-based filesystem (such as tmpfs)
        can massively speed up the test suite.

--chain-lint::
--no-chain-lint::
        If --chain-lint is enabled, the test harness will check each
        test to make sure that it properly "&&-chains" all commands (so
        that a failure in the middle does not go unnoticed by the final
        exit code of the test). This check is performed in addition to
        running the tests themselves. You may also enable or disable
        this feature by setting the GIT_TEST_CHAIN_LINT environment
        variable to "1" or "0", respectively.

        A few test scripts disable some of the more advanced
        chain-linting detection in the name of efficiency. You can
        override this by setting the GIT_TEST_CHAIN_LINT_HARDER
        environment variable to "1".

--stress::
        Run the test script repeatedly in multiple parallel jobs until
        one of them fails.  Useful for reproducing rare failures in
        flaky tests.  The number of parallel jobs is, in order of
        precedence: the value of the GIT_TEST_STRESS_LOAD
        environment variable, or twice the number of available
        processors (as shown by the 'getconf' utility), or 8.
        Implies `--verbose -x --immediate` to get the most information
        about the failure.  Note that the verbose output of each test
        job is saved to 't/test-results/$TEST_NAME.stress-<nr>.out',
        and only the output of the failed test job is shown on the
        terminal.  The names of the trash directories get a
        '.stress-<nr>' suffix, and the trash directory of the failed
        test job is renamed to end with a '.stress-failed' suffix.

--stress-jobs=<N>::
        Override the number of parallel jobs. Implies `--stress`.

--stress-limit=<N>::
        When combined with --stress run the test script repeatedly
        this many times in each of the parallel jobs or until one of
        them fails, whichever comes first. Implies `--stress`.

You can also set the GIT_TEST_INSTALLED environment variable to
the bindir of an existing git installation to test that installation.
You still need to have built this git sandbox, from which various
test-* support programs, templates, and perl libraries are used.
If your installed git is incomplete, it will silently test parts of
your built version instead.

When using GIT_TEST_INSTALLED, you can also set GIT_TEST_EXEC_PATH to
override the location of the dashed-form subcommands (what
GIT_EXEC_PATH would be used for during normal operation).
GIT_TEST_EXEC_PATH defaults to `$GIT_TEST_INSTALLED/git --exec-path`.


Skipping Tests
--------------

In some environments, certain tests have no way of succeeding
due to platform limitation, such as lack of 'unzip' program, or
filesystem that do not allow arbitrary sequence of non-NUL bytes
as pathnames.

You should be able to say something like

    $ GIT_SKIP_TESTS=t9200.8 sh ./t9200-git-cvsexport-commit.sh

and even:

    $ GIT_SKIP_TESTS='t[0-4]??? t91?? t9200.8' make

to omit such tests.  The value of the environment variable is a
SP separated list of patterns that tells which tests to skip,
and either can match the "t[0-9]{4}" part to skip the whole
test, or t[0-9]{4} followed by ".$number" to say which
particular test to skip.

For an individual test suite --run could be used to specify that
only some tests should be run or that some tests should be
excluded from a run.

The argument for --run, <test-selector>, is a list of description
substrings or globs or individual test numbers or ranges with an
optional negation prefix (of '!') that define what tests in a test
suite to include (or exclude, if negated) in the run.  A range is two
numbers separated with a dash and matches a range of tests with both
ends been included.  You may omit the first or the second number to
mean "from the first test" or "up to the very last test" respectively.

The argument to --run is split on commas into separate strings,
numbers, and ranges, and picks all tests that match any of the
individual selection criteria.  If the substring of the description
text that you want to match includes a comma, use the glob character
'?' instead.  For example --run='rebase,merge?cherry-pick' would match
on all tests that match either the glob *rebase* or the glob
*merge?cherry-pick*.

If --run starts with an unprefixed number or range the initial
set of tests to run is empty. If the first item starts with '!'
all the tests are added to the initial set.  After initial set is
determined every test number or range is added or excluded from
the set one by one, from left to right.

For example, to run only tests up to a specific test (21), one
could do this:

    $ sh ./t9200-git-cvsexport-commit.sh --run='1-21'

or this:

    $ sh ./t9200-git-cvsexport-commit.sh --run='-21'

Common case is to run several setup tests (1, 2, 3) and then a
specific test (21) that relies on that setup:

    $ sh ./t9200-git-cvsexport-commit.sh --run='1,2,3,21'

or:

    $ sh ./t9200-git-cvsexport-commit.sh --run=1,2,3,21

or:

    $ sh ./t9200-git-cvsexport-commit.sh --run='-3,21'

As noted above, the test set is built by going through the items
from left to right, so this:

    $ sh ./t9200-git-cvsexport-commit.sh --run='1-4,!3'

will run tests 1, 2, and 4.  Items that come later have higher
precedence.  It means that this:

    $ sh ./t9200-git-cvsexport-commit.sh --run='!3,1-4'

would just run tests from 1 to 4, including 3.

You may use negation with ranges.  The following will run all
test in the test suite except from 7 up to 11:

    $ sh ./t9200-git-cvsexport-commit.sh --run='!7-11'

Sometimes there may be multiple tests with e.g. "setup" in their name
that are needed and rather than figuring out the number for all of them
we can just use "setup" as a substring/glob to match against the test
description:

    $ sh ./t0050-filesystem.sh --run=setup,9-11

or one could select both the setup tests and the rename ones (assuming all
relevant tests had those words in their descriptions):

    $ sh ./t0050-filesystem.sh --run=setup,rename

Some tests in a test suite rely on the previous tests performing
certain actions, specifically some tests are designated as
"setup" test, so you cannot _arbitrarily_ disable one test and
expect the rest to function correctly.

--run is mostly useful when you want to focus on a specific test
and know what setup is needed for it.  Or when you want to run
everything up to a certain test.


Running tests with special setups
---------------------------------

The whole test suite could be run to test some special features
that cannot be easily covered by a few specific test cases. These
could be enabled by running the test suite with correct GIT_TEST_
environment set.

GIT_TEST_FAIL_PREREQS=<boolean> fails all prerequisites. This is
useful for discovering issues with the tests where say a later test
implicitly depends on an optional earlier test.

There's a "FAIL_PREREQS" prerequisite that can be used to test for
whether this mode is active, and e.g. skip some tests that are hard to
refactor to deal with it. The "SYMLINKS" prerequisite is currently
excluded as so much relies on it, but this might change in the future.

GIT_TEST_SPLIT_INDEX=<boolean> forces split-index mode on the whole
test suite. Accept any boolean values that are accepted by git-config.

GIT_TEST_PASSING_SANITIZE_LEAK=true skips those tests that haven't
declared themselves as leak-free by setting
"TEST_PASSES_SANITIZE_LEAK=true" before sourcing "test-lib.sh". This
test mode is used by the "linux-leaks" CI target.

GIT_TEST_PASSING_SANITIZE_LEAK=check checks that our
"TEST_PASSES_SANITIZE_LEAK=true" markings are current. Rather than
skipping those tests that haven't set "TEST_PASSES_SANITIZE_LEAK=true"
before sourcing "test-lib.sh" this mode runs them with
"--invert-exit-code". This is used to check that there's a one-to-one
mapping between "TEST_PASSES_SANITIZE_LEAK=true" and those tests that
pass under "SANITIZE=leak". This is especially useful when testing a
series that fixes various memory leaks with "git rebase -x".

GIT_TEST_SANITIZE_LEAK_LOG=true will log memory leaks to
"test-results/$TEST_NAME.leak/trace.*" files. The logs include a
"dedup_token" (see +"ASAN_OPTIONS=help=1 ./git") and other options to
make logs +machine-readable.

GIT_TEST_PASSING_SANITIZE_LEAK=check when combined with "--immediate"
will run to completion faster, and result in the same failing
tests. The only practical reason to run
GIT_TEST_PASSING_SANITIZE_LEAK=check without "--immediate" is to
combine it with "GIT_TEST_SANITIZE_LEAK_LOG=true". If we stop at the
first failing test case our leak logs won't show subsequent leaks we
might have run into.

GIT_TEST_PROTOCOL_VERSION=<n>, when set, makes 'protocol.version'
default to n.

GIT_TEST_FULL_IN_PACK_ARRAY=<boolean> exercises the uncommon
pack-objects code path where there are more than 1024 packs even if
the actual number of packs in repository is below this limit. Accept
any boolean values that are accepted by git-config.

GIT_TEST_OE_SIZE=<n> exercises the uncommon pack-objects code path
where we do not cache object size in memory and read it from existing
packs on demand. This normally only happens when the object size is
over 2GB. This variable forces the code path on any object larger than
<n> bytes.

GIT_TEST_OE_DELTA_SIZE=<n> exercises the uncommon pack-objects code
path where deltas larger than this limit require extra memory
allocation for bookkeeping.

GIT_TEST_VALIDATE_INDEX_CACHE_ENTRIES=<boolean> checks that cache-tree
records are valid when the index is written out or after a merge. This
is mostly to catch missing invalidation. Default is true.

GIT_TEST_COMMIT_GRAPH=<boolean>, when true, forces the commit-graph to
be written after every 'git commit' command, and overrides the
'core.commitGraph' setting to true.

GIT_TEST_COMMIT_GRAPH_CHANGED_PATHS=<boolean>, when true, forces
commit-graph write to compute and write changed path Bloom filters for
every 'git commit-graph write', as if the `--changed-paths` option was
passed in.

GIT_TEST_FSMONITOR=$PWD/t7519/fsmonitor-all exercises the fsmonitor
code paths for utilizing a (hook based) file system monitor to speed up
detecting new or changed files.

GIT_TEST_INDEX_VERSION=<n> exercises the index read/write code path
for the index version specified.  Can be set to any valid version
(currently 2, 3, or 4).

GIT_TEST_PACK_SPARSE=<boolean> if disabled will default the pack-objects
builtin to use the non-sparse object walk. This can still be overridden by
the --sparse command-line argument.

GIT_TEST_PRELOAD_INDEX=<boolean> exercises the preload-index code path
by overriding the minimum number of cache entries required per thread.

GIT_TEST_ADD_I_USE_BUILTIN=<boolean>, when false, disables the
built-in version of git add -i. See 'add.interactive.useBuiltin' in
git-config(1).

GIT_TEST_INDEX_THREADS=<n> enables exercising the multi-threaded loading
of the index for the whole test suite by bypassing the default number of
cache entries and thread minimums. Setting this to 1 will make the
index loading single threaded.

GIT_TEST_MULTI_PACK_INDEX=<boolean>, when true, forces the multi-pack-
index to be written after every 'git repack' command, and overrides the
'core.multiPackIndex' setting to true.

GIT_TEST_MULTI_PACK_INDEX_WRITE_BITMAP=<boolean>, when true, sets the
'--bitmap' option on all invocations of 'git multi-pack-index write',
and ignores pack-objects' '--write-bitmap-index'.

GIT_TEST_SIDEBAND_ALL=<boolean>, when true, overrides the
'uploadpack.allowSidebandAll' setting to true, and when false, forces
fetch-pack to not request sideband-all (even if the server advertises
sideband-all).

GIT_TEST_DISALLOW_ABBREVIATED_OPTIONS=<boolean>, when true (which is
the default when running tests), errors out when an abbreviated option
is used.

GIT_TEST_DEFAULT_HASH=<hash-algo> specifies which hash algorithm to
use in the test scripts. Recognized values for <hash-algo> are "sha1"
and "sha256".

GIT_TEST_WRITE_REV_INDEX=<boolean>, when true enables the
'pack.writeReverseIndex' setting.

GIT_TEST_SPARSE_INDEX=<boolean>, when true enables index writes to use the
sparse-index format by default.

GIT_TEST_CHECKOUT_WORKERS=<n> overrides the 'checkout.workers' setting
to <n> and 'checkout.thresholdForParallelism' to 0, forcing the
execution of the parallel-checkout code.

GIT_TEST_FATAL_REGISTER_SUBMODULE_ODB=<boolean>, when true, makes
registering submodule ODBs as alternates a fatal action. Support for
this environment variable can be removed once the migration to
explicitly providing repositories when accessing submodule objects is
complete or needs to be abandoned for whatever reason (in which case the
migrated codepaths still retain their performance benefits).

GIT_TEST_REQUIRE_PREREQ=<list> allows specifying a space separated list of
prereqs that are required to succeed. If a prereq in this list is triggered by
a test and then fails then the whole test run will abort. This can help to make
sure the expected tests are executed and not silently skipped when their
dependency breaks or is simply not present in a new environment.

Naming Tests
------------

The test files are named as:

        tNNNN-commandname-details.sh

where N is a decimal digit.

First digit tells the family:

        0 - the absolute basics and global stuff
        1 - the basic commands concerning database
        2 - the basic commands concerning the working tree
        3 - the other basic commands (e.g. ls-files)
        4 - the diff commands
        5 - the pull and exporting commands
        6 - the revision tree commands (even e.g. merge-base)
        7 - the porcelainish commands concerning the working tree
        8 - the porcelainish commands concerning forensics
        9 - the git tools

Second digit tells the particular command we are testing.

Third digit (optionally) tells the particular switch or group of switches
we are testing.

If you create files under t/ directory (i.e. here) that is not
the top-level test script, never name the file to match the above
pattern.  The Makefile here considers all such files as the
top-level test script and tries to run all of them.  Care is
especially needed if you are creating a common test library
file, similar to test-lib.sh, because such a library file may
not be suitable for standalone execution.


Writing Tests
-------------

The test script is written as a shell script.  It should start
with the standard "#!/bin/sh", and an
assignment to variable 'test_description', like this:

        #!/bin/sh

        test_description='xxx test (option --frotz)

        This test registers the following structure in the cache
        and tries to run git-ls-files with option --frotz.'


Source 'test-lib.sh'
--------------------

After assigning test_description, the test script should source
test-lib.sh like this:

        . ./test-lib.sh

This test harness library does the following things:

 - If the script is invoked with command line argument --help
   (or -h), it shows the test_description and exits.

 - Creates an empty test directory with an empty .git/objects database
   and chdir(2) into it.  This directory is 't/trash
   directory.$test_name_without_dotsh', with t/ subject to change by
   the --root option documented above, and a '.stress-<N>' suffix
   appended by the --stress option.

 - Defines standard test helper functions for your scripts to
   use.  These functions are designed to make all scripts behave
   consistently when command line arguments --verbose (or -v),
   --debug (or -d), and --immediate (or -i) is given.

Recommended style
-----------------
Here are some recommented styles when writing test case.

 - Keep test title the same line with test helper function itself.

   Take test_expect_success helper for example, write it like:

  test_expect_success 'test title' '
      ... test body ...
  '

   Instead of:

  test_expect_success \
      'test title' \
      '... test body ...'


 - End the line with a single quote.

 - Indent the body of here-document, and use "<<-" instead of "<<"
   to strip leading TABs used for indentation:

  test_expect_success 'test something' '
      cat >expect <<-\EOF &&
      one
      two
      three
      EOF
      test_something > actual &&
      test_cmp expect actual
  '

   Instead of:

  test_expect_success 'test something' '
      cat >expect <<\EOF &&
  one
  two
  three
  EOF
      test_something > actual &&
      test_cmp expect actual
  '

 - Quote or escape the EOF delimiter that begins a here-document if
   there is no parameter and other expansion in it, to signal readers
   that they can skim it more casually:

  cmd <<-\EOF
  literal here-document text without any expansion
  EOF


Do's & don'ts
-------------

Here are a few examples of things you probably should and shouldn't do
when writing tests.

Here are the "do's:"

 - Put all code inside test_expect_success and other assertions.

   Even code that isn't a test per se, but merely some setup code
   should be inside a test assertion.

 - Chain your test assertions

   Write test code like this:

        git merge foo &&
        git push bar &&
        test ...

   Instead of:

        git merge hla
        git push gh
        test ...

   That way all of the commands in your tests will succeed or fail. If
   you must ignore the return value of something, consider using a
   helper function (e.g. use sane_unset instead of unset, in order
   to avoid unportable return value for unsetting a variable that was
   already unset), or prepending the command with test_might_fail or
   test_must_fail.

 - Check the test coverage for your tests. See the "Test coverage"
   below.

   Don't blindly follow test coverage metrics; if a new function you added
   doesn't have any coverage, then you're probably doing something wrong,
   but having 100% coverage doesn't necessarily mean that you tested
   everything.

   Tests that are likely to smoke out future regressions are better
   than tests that just inflate the coverage metrics.

 - When a test checks for an absolute path that a git command generated,
   construct the expected value using $(pwd) rather than $PWD,
   $TEST_DIRECTORY, or $TRASH_DIRECTORY. It makes a difference on
   Windows, where the shell (MSYS bash) mangles absolute path names.
   For details, see the commit message of 4114156ae9.

 - Remember that inside the <script> part, the standard output and
   standard error streams are discarded, and the test harness only
   reports "ok" or "not ok" to the end user running the tests. Under
   --verbose, they are shown to help debug the tests.

 - Be careful when you loop

   You may need to verify multiple things in a loop, but the
   following does not work correctly:

        test_expect_success 'test three things' '
            for i in one two three
            do
                test_something "$i"
            done &&
            test_something_else
        '

   Because the status of the loop itself is the exit status of the
   test_something in the last round, the loop does not fail when
   "test_something" for "one" or "two" fails.  This is not what you
   want.

   Instead, you can break out of the loop immediately when you see a
   failure.  Because all test_expect_* snippets are executed inside
   a function, "return 1" can be used to fail the test immediately
   upon a failure:

        test_expect_success 'test three things' '
            for i in one two three
            do
                test_something "$i" || return 1
            done &&
            test_something_else
        '

   Note that we still &&-chain the loop to propagate failures from
   earlier commands.


And here are the "don'ts:"

 - Don't exit() within a <script> part.

   The harness will catch this as a programming error of the test.
   Use test_done instead if you need to stop the tests early (see
   "Skipping tests" below).

 - Don't use '! git cmd' when you want to make sure the git command
   exits with failure in a controlled way by calling "die()".  Instead,
   use 'test_must_fail git cmd'.  This will signal a failure if git
   dies in an unexpected way (e.g. segfault).

   On the other hand, don't use test_must_fail for running regular
   platform commands; just use '! cmd'.  We are not in the business
   of verifying that the world given to us sanely works.

 - Don't feed the output of a git command to a pipe, as in:

     git -C repo ls-files |
     xargs -n 1 basename |
     grep foo

   which will discard git's exit code and may mask a crash. In the
   above example, all exit codes are ignored except grep's.

   Instead, write the output of that command to a temporary
   file with ">" or assign it to a variable with "x=$(git ...)" rather
   than pipe it.

 - Don't use command substitution in a way that discards git's exit
   code. When assigning to a variable, the exit code is not discarded,
   e.g.:

     x=$(git cat-file -p $sha) &&
     ...

   is OK because a crash in "git cat-file" will cause the "&&" chain
   to fail, but:

     test "refs/heads/foo" = "$(git symbolic-ref HEAD)"

   is not OK and a crash in git could go undetected.

 - Don't use perl without spelling it as "$PERL_PATH". This is to help
   our friends on Windows where the platform Perl often adds CR before
   the end of line, and they bundle Git with a version of Perl that
   does not do so, whose path is specified with $PERL_PATH. Note that we
   provide a "perl" function which uses $PERL_PATH under the hood, so
   you do not need to worry when simply running perl in the test scripts
   (but you do, for example, on a shebang line or in a sub script
   created via "write_script").

 - Don't use sh without spelling it as "$SHELL_PATH", when the script
   can be misinterpreted by broken platform shell (e.g. Solaris).

 - Don't chdir around in tests.  It is not sufficient to chdir to
   somewhere and then chdir back to the original location later in
   the test, as any intermediate step can fail and abort the test,
   causing the next test to start in an unexpected directory.  Do so
   inside a subshell if necessary.

 - Don't save and verify the standard error of compound commands, i.e.
   group commands, subshells, and shell functions (except test helper
   functions like 'test_must_fail') like this:

     ( cd dir && git cmd ) 2>error &&
     test_cmp expect error

   When running the test with '-x' tracing, then the trace of commands
   executed in the compound command will be included in standard error
   as well, quite possibly throwing off the subsequent checks examining
   the output.  Instead, save only the relevant git command's standard
   error:

     ( cd dir && git cmd 2>../error ) &&
     test_cmp expect error

 - Don't break the TAP output

   The raw output from your test may be interpreted by a TAP harness. TAP
   harnesses will ignore everything they don't know about, but don't step
   on their toes in these areas:

   - Don't print lines like "$x..$y" where $x and $y are integers.

   - Don't print lines that begin with "ok" or "not ok".

   TAP harnesses expect a line that begins with either "ok" and "not
   ok" to signal a test passed or failed (and our harness already
   produces such lines), so your script shouldn't emit such lines to
   their output.

   You can glean some further possible issues from the TAP grammar
   (see https://metacpan.org/pod/TAP::Parser::Grammar#TAP-GRAMMAR)
   but the best indication is to just run the tests with prove(1),
   it'll complain if anything is amiss.


Skipping tests
--------------

If you need to skip tests you should do so by using the three-arg form
of the test_* functions (see the "Test harness library" section
below), e.g.:

    test_expect_success PERL 'I need Perl' '
        perl -e "hlagh() if unf_unf()"
    '

The advantage of skipping tests like this is that platforms that don't
have the PERL and other optional dependencies get an indication of how
many tests they're missing.

If the test code is too hairy for that (i.e. does a lot of setup work
outside test assertions) you can also skip all remaining tests by
setting skip_all and immediately call test_done:

        if ! test_have_prereq PERL
        then
            skip_all='skipping perl interface tests, perl not available'
            test_done
        fi

The string you give to skip_all will be used as an explanation for why
the test was skipped.

End with test_done
------------------

Your script will be a sequence of tests, using helper functions
from the test harness library.  At the end of the script, call
'test_done'.


Test harness library
--------------------

There are a handful helper functions defined in the test harness
library for your script to use. Some of them are listed below;
see test-lib-functions.sh for the full list and their options.

 - test_expect_success [<prereq>] <message> <script>

   Usually takes two strings as parameters, and evaluates the
   <script>.  If it yields success, test is considered
   successful.  <message> should state what it is testing.

   Example:

        test_expect_success \
            'git-write-tree should be able to write an empty tree.' \
            'tree=$(git-write-tree)'

   If you supply three parameters the first will be taken to be a
   prerequisite; see the test_set_prereq and test_have_prereq
   documentation below:

        test_expect_success TTY 'git --paginate rev-list uses a pager' \
            ' ... '

   You can also supply a comma-separated list of prerequisites, in the
   rare case where your test depends on more than one:

        test_expect_success PERL,PYTHON 'yo dawg' \
            ' test $(perl -E 'print eval "1 +" . qx[python -c "print 2"]') == "4" '

 - test_expect_failure [<prereq>] <message> <script>

   This is NOT the opposite of test_expect_success, but is used
   to mark a test that demonstrates a known breakage.  Unlike
   the usual test_expect_success tests, which say "ok" on
   success and "FAIL" on failure, this will say "FIXED" on
   success and "still broken" on failure.  Failures from these
   tests won't cause -i (immediate) to stop.

   Like test_expect_success this function can optionally use a three
   argument invocation with a prerequisite as the first argument.

 - test_debug <script>

   This takes a single argument, <script>, and evaluates it only
   when the test script is started with --debug command line
   argument.  This is primarily meant for use during the
   development of a new test script.

 - debug [options] <git-command>

   Run a git command inside a debugger. This is primarily meant for
   use when debugging a failing test script. With '-t', use your
   original TERM instead of test-lib.sh's "dumb", so that your
   debugger interface has colors.

 - test_done

   Your test script must have test_done at the end.  Its purpose
   is to summarize successes and failures in the test script and
   exit with an appropriate error code.

 - test_tick

   Make commit and tag names consistent by setting the author and
   committer times to defined state.  Subsequent calls will
   advance the times by a fixed amount.

 - test_commit <message> [<filename> [<contents>]]

   Creates a commit with the given message, committing the given
   file with the given contents (default for both is to reuse the
   message string), and adds a tag (again reusing the message
   string as name).  Calls test_tick to make the SHA-1s
   reproducible.

 - test_merge <message> <commit-or-tag>

   Merges the given rev using the given message.  Like test_commit,
   creates a tag and calls test_tick before committing.

 - test_set_prereq <prereq>

   Set a test prerequisite to be used later with test_have_prereq. The
   test-lib will set some prerequisites for you, see the
   "Prerequisites" section below for a full list of these.

   Others you can set yourself and use later with either
   test_have_prereq directly, or the three argument invocation of
   test_expect_success and test_expect_failure.

 - test_have_prereq <prereq>

   Check if we have a prerequisite previously set with test_set_prereq.
   The most common way to use this explicitly (as opposed to the
   implicit use when an argument is passed to test_expect_*) is to skip
   all the tests at the start of the test script if we don't have some
   essential prerequisite:

        if ! test_have_prereq PERL
        then
            skip_all='skipping perl interface tests, perl not available'
            test_done
        fi

 - test_expect_code <exit-code> <command>

   Run a command and ensure that it exits with the given exit code.
   For example:

        test_expect_success 'Merge with d/f conflicts' '
                test_expect_code 1 git merge "merge msg" B master
        '

 - test_must_fail [<options>] <git-command>

   Run a git command and ensure it fails in a controlled way.  Use
   this instead of "! <git-command>".  When git-command dies due to a
   segfault, test_must_fail diagnoses it as an error; "! <git-command>"
   treats it as just another expected failure, which would let such a
   bug go unnoticed.

   Accepts the following options:

     ok=<signal-name>[,<...>]:
       Don't treat an exit caused by the given signal as error.
       Multiple signals can be specified as a comma separated list.
       Currently recognized signal names are: sigpipe, success.
       (Don't use 'success', use 'test_might_fail' instead.)

 - test_might_fail [<options>] <git-command>

   Similar to test_must_fail, but tolerate success, too.  Use this
   instead of "<git-command> || :" to catch failures due to segv.

   Accepts the same options as test_must_fail.

 - test_cmp <expected> <actual>

   Check whether the content of the <actual> file matches the
   <expected> file.  This behaves like "cmp" but produces more
   helpful output when the test is run with "-v" option.

 - test_cmp_rev <expected> <actual>

   Check whether the <expected> rev points to the same commit as the
   <actual> rev.

 - test_line_count (= | -lt | -ge | ...) <length> <file>

   Check whether a file has the length it is expected to.

 - test_path_is_file <path>
   test_path_is_dir <path>
   test_path_is_missing <path>

   Check if the named path is a file, if the named path is a
   directory, or if the named path does not exist, respectively,
   and fail otherwise.

 - test_when_finished <script>

   Prepend <script> to a list of commands to run to clean up
   at the end of the current test.  If some clean-up command
   fails, the test will not pass.

   Example:

        test_expect_success 'branch pointing to non-commit' '
                git rev-parse HEAD^{tree} >.git/refs/heads/invalid &&
                test_when_finished "git update-ref -d refs/heads/invalid" &&
                ...
        '

 - test_atexit <script>

   Prepend <script> to a list of commands to run unconditionally to
   clean up before the test script exits, e.g. to stop a daemon:

        test_expect_success 'test git daemon' '
                git daemon &
                daemon_pid=$! &&
                test_atexit 'kill $daemon_pid' &&
                hello world
        '

   The commands will be executed before the trash directory is removed,
   i.e. the atexit commands will still be able to access any pidfiles or
   socket files.

   Note that these commands will be run even when a test script run
   with '--immediate' fails.  Be careful with your atexit commands to
   minimize any changes to the failed state.

 - test_write_lines <lines>

   Write <lines> on standard output, one line per argument.
   Useful to prepare multi-line files in a compact form.

   Example:

        test_write_lines a b c d e f g >foo

   Is a more compact equivalent of:
        cat >foo <<-EOF
        a
        b
        c
        d
        e
        f
        g
        EOF


 - test_pause [options]

        This command is useful for writing and debugging tests and must be
        removed before submitting. It halts the execution of the test and
        spawns a shell in the trash directory. Exit the shell to continue
        the test. Example:

        test_expect_success 'test' '
                git do-something >actual &&
                test_pause &&
                test_cmp expected actual
        '

 - test_ln_s_add <path1> <path2>

   This function helps systems whose filesystem does not support symbolic
   links. Use it to add a symbolic link entry to the index when it is not
   important that the file system entry is a symbolic link, i.e., instead
   of the sequence

        ln -s foo bar &&
        git add bar

   Sometimes it is possible to split a test in a part that does not need
   the symbolic link in the file system and a part that does; then only
   the latter part need be protected by a SYMLINKS prerequisite (see below).

 - test_oid_init

   This function loads facts and useful object IDs related to the hash
   algorithm(s) in use from the files in t/oid-info.

 - test_oid_cache

   This function reads per-hash algorithm information from standard
   input (usually a heredoc) in the format described in
   t/oid-info/README.  This is useful for test-specific values, such as
   object IDs, which must vary based on the hash algorithm.

   Certain fixed values, such as hash sizes and common placeholder
   object IDs, can be loaded with test_oid_init (described above).

 - test_oid <key>

   This function looks up a value for the hash algorithm in use, based
   on the key given.  The value must have been loaded using
   test_oid_init or test_oid_cache.  Providing an unknown key is an
   error.

 - yes [<string>]

   This is often seen in modern UNIX but some platforms lack it, so
   the test harness overrides the platform implementation with a
   more limited one.  Use this only when feeding a handful lines of
   output to the downstream---unlike the real version, it generates
   only up to 99 lines.

 - test_bool_env <env-variable-name> <default-value>

   Given the name of an environment variable with a bool value,
   normalize its value to a 0 (true) or 1 (false or empty string)
   return code.  Return with code corresponding to the given default
   value if the variable is unset.
   Abort the test script if either the value of the variable or the
   default are not valid bool values.


Prerequisites
-------------

These are the prerequisites that the test library predefines with
test_have_prereq.

See the prereq argument to the test_* functions in the "Test harness
library" section above and the "test_have_prereq" function for how to
use these, and "test_set_prereq" for how to define your own.

 - PYTHON

   Git wasn't compiled with NO_PYTHON=YesPlease. Wrap any tests that
   need Python with this.

 - PERL

   Git wasn't compiled with NO_PERL=YesPlease.

   Even without the PERL prerequisite, tests can assume there is a
   usable perl interpreter at $PERL_PATH, though it need not be
   particularly modern.

 - POSIXPERM

   The filesystem supports POSIX style permission bits.

 - BSLASHPSPEC

   Backslashes in pathspec are not directory separators. This is not
   set on Windows. See 6fd1106a for details.

 - EXECKEEPSPID

   The process retains the same pid across exec(2). See fb9a2bea for
   details.

 - PIPE

   The filesystem we're on supports creation of FIFOs (named pipes)
   via mkfifo(1).

 - SYMLINKS

   The filesystem we're on supports symbolic links. E.g. a FAT
   filesystem doesn't support these. See 704a3143 for details.

 - SANITY

   Test is not run by root user, and an attempt to write to an
   unwritable file is expected to fail correctly.

 - PCRE

   Git was compiled with support for PCRE. Wrap any tests
   that use git-grep --perl-regexp or git-grep -P in these.

 - CASE_INSENSITIVE_FS

   Test is run on a case insensitive file system.

 - UTF8_NFD_TO_NFC

   Test is run on a filesystem which converts decomposed utf-8 (nfd)
   to precomposed utf-8 (nfc).

 - PTHREADS

   Git wasn't compiled with NO_PTHREADS=YesPlease.

 - REFFILES

   Test is specific to packed/loose ref storage, and should be
   disabled for other ref storage backends


Tips for Writing Tests
----------------------

As with any programming projects, existing programs are the best
source of the information.  However, do _not_ emulate
t0000-basic.sh when writing your tests.  The test is special in
that it tries to validate the very core of Git.  For example, it
knows that there will be 256 subdirectories under .git/objects/,
and it knows that the object ID of an empty tree is a certain
40-byte string.  This is deliberately done so in t0000-basic.sh
because the things the very basic core test tries to achieve is
to serve as a basis for people who are changing the Git internals
drastically.  For these people, after making certain changes,
not seeing failures from the basic test _is_ a failure.  And
such drastic changes to the core Git that even changes these
otherwise supposedly stable object IDs should be accompanied by
an update to t0000-basic.sh.

However, other tests that simply rely on basic parts of the core
Git working properly should not have that level of intimate
knowledge of the core Git internals.  If all the test scripts
hardcoded the object IDs like t0000-basic.sh does, that defeats
the purpose of t0000-basic.sh, which is to isolate that level of
validation in one place.  Your test also ends up needing
updating when such a change to the internal happens, so do _not_
do it and leave the low level of validation to t0000-basic.sh.

Test coverage
-------------

You can use the coverage tests to find code paths that are not being
used or properly exercised yet.

To do that, run the coverage target at the top-level (not in the t/
directory):

    make coverage

That'll compile Git with GCC's coverage arguments, and generate a test
report with gcov after the tests finish. Running the coverage tests
can take a while, since running the tests in parallel is incompatible
with GCC's coverage mode.

After the tests have run you can generate a list of untested
functions:

    make coverage-untested-functions

You can also generate a detailed per-file HTML report using the
Devel::Cover module. To install it do:

   # On Debian or Ubuntu:
   sudo aptitude install libdevel-cover-perl

   # From the CPAN with cpanminus
   curl -L http://cpanmin.us | perl - --sudo --self-upgrade
   cpanm --sudo Devel::Cover

Then, at the top-level:

    make cover_db_html

That'll generate a detailed cover report in the "cover_db_html"
directory, which you can then copy to a webserver, or inspect locally
in a browser.