Calculating duration from a single uint64_t is simpler than from a struct
timeval. Change throughput measurement from gettimeofday() to
getnanotime().
Also calculate misec only if needed, and change integer division to integer
multiplication + shift, which should be slightly faster.
Signed-off-by: Karsten Blees <blees@dcon.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Calculating duration from a single uint64_t is simpler than from a struct
timeval. Change performance measurement for 'advice.statusuoption' from
gettimeofday() to getnanotime().
Also initialize t_begin to prevent uninitialized variable warning.
Signed-off-by: Karsten Blees <blees@dcon.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Use trace_performance to measure and print execution time and command line
arguments of the entire main() function. In constrast to the shell's 'time'
utility, which measures total time of the parent process, this logs all
involved git commands recursively. This is particularly useful to debug
performance issues of scripted commands (i.e. which git commands were
called with which parameters, and how long did they execute).
Due to git's deliberate use of exit(), the implementation uses an atexit
routine rather than just adding trace_performance_since() at the end of
main().
Usage example: > GIT_TRACE_PERFORMANCE=~/git-trace.log git stash list
Creates a log file like this:
23:57:38.638765 trace.c:405 performance: 0.000310107 s: git command: 'git' 'rev-parse' '--git-dir'
23:57:38.644387 trace.c:405 performance: 0.000261759 s: git command: 'git' 'rev-parse' '--show-toplevel'
23:57:38.646207 trace.c:405 performance: 0.000304468 s: git command: 'git' 'config' '--get-colorbool' 'color.interactive'
23:57:38.648491 trace.c:405 performance: 0.000241667 s: git command: 'git' 'config' '--get-color' 'color.interactive.help' 'red bold'
23:57:38.650465 trace.c:405 performance: 0.000243063 s: git command: 'git' 'config' '--get-color' '' 'reset'
23:57:38.654850 trace.c:405 performance: 0.025126313 s: git command: 'git' 'stash' 'list'
Signed-off-by: Karsten Blees <blees@dcon.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Add trace_performance and trace_performance_since macros that print a
duration and an optional printf-formatted text to the file specified in
environment variable GIT_TRACE_PERFORMANCE.
These macros, in conjunction with getnanotime(), are intended to simplify
performance measurements from within the application (i.e. profiling via
manual instrumentation, rather than using an external profiling tool).
Unless enabled via GIT_TRACE_PERFORMANCE, these macros have no noticeable
impact on performance, so that test code for well known time killers may
be shipped in release builds. Alternatively, a developer could provide an
additional performance patch (not meant for master) that allows reviewers
to reproduce performance tests more easily, e.g. on other platforms or
using their own repositories.
Usage examples:
Simple use case (measure one code section):
uint64_t start = getnanotime();
/* code section to measure */
trace_performance_since(start, "foobar");
Complex use case (measure repetitive code sections):
uint64_t t = 0;
for (;;) {
/* ignore */
t -= getnanotime();
/* code section to measure */
t += getnanotime();
/* ignore */
}
trace_performance(t, "frotz");
Signed-off-by: Karsten Blees <blees@dcon.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Add a getnanotime() function that returns nanoseconds since 01/01/1970 as
unsigned 64-bit integer (i.e. overflows in july 2554). This is easier to
work with than e.g. struct timeval or struct timespec. Basing the timer on
the epoch allows using the results with other time-related APIs.
To simplify adaption to different platforms, split the implementation into
a common getnanotime() and a platform-specific highres_nanos() function.
The common getnanotime() function handles errors, falling back to
gettimeofday() if highres_nanos() isn't implemented or doesn't work.
getnanotime() is also responsible for normalizing to the epoch. The offset
to the system clock is calculated only once on initialization, i.e.
manually setting the system clock has no impact on the timer (except if
the fallback gettimeofday() is in use). Git processes are typically short
lived, so we don't need to handle clock drift.
The highres_nanos() function returns monotonically increasing nanoseconds
relative to some arbitrary point in time (e.g. system boot), or 0 on
failure. Providing platform-specific implementations should be relatively
easy, e.g. adapting to clock_gettime() as defined by the POSIX realtime
extensions is seven lines of code.
This version includes highres_nanos() implementations for:
* Linux: using clock_gettime(CLOCK_MONOTONIC)
* Windows: using QueryPerformanceCounter()
Todo:
* enable clock_gettime() on more platforms
* add Mac OSX version, e.g. using mach_absolute_time + mach_timebase_info
Signed-off-by: Karsten Blees <blees@dcon.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This is useful to see where trace output came from.
Add 'const char *file, int line' parameters to the printing functions and
rename them to *_fl.
Add trace_printf* and trace_strbuf macros resolving to the *_fl functions
and let the preprocessor fill in __FILE__ and __LINE__.
As the trace_printf* functions take a variable number of arguments, this
requires variadic macros (i.e. '#define foo(...) foo_impl(__VA_ARGS__)'.
Though part of C99, it is unclear whether older compilers support this.
Thus keep the old functions and only enable variadic macros for GNUC and
MSVC 2005+ (_MSC_VER 1400). This has the nice side effect that the old
C-style declarations serve as documentation how the macros are to be used.
Print 'file:line ' as prefix to each trace line. Align the remaining trace
output at column 40 to accommodate 18 char file names + 4 digit line
number (currently there are 30 *.c files of length 18 and just 11 of 19).
Trace output from longer source files (e.g. builtin/receive-pack.c) will
not be aligned.
Signed-off-by: Karsten Blees <blees@dcon.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
No functional changes, just move stuff around so that the next patch isn't
that ugly...
Signed-off-by: Karsten Blees <blees@dcon.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This is useful to tell apart trace output of separate test runs.
It can also be used for basic, coarse-grained performance analysis. Note
that the accuracy is tainted by writing to the trace file, and you have to
calculate the deltas yourself (which is next to impossible if multiple
threads or processes are involved).
Signed-off-by: Karsten Blees <blees@dcon.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Some unit-tests use trace output to verify internal state, and unstable
output such as timestamps and line numbers are not useful there.
Disable additional trace output if GIT_TRACE_BARE is set.
Signed-off-by: Karsten Blees <blees@dcon.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
To be able to add a common prefix or suffix to all trace output (e.g.
a timestamp or file:line of the caller), factor out common setup and
cleanup tasks of the trace* functions.
When adding a common prefix, it makes sense that the output of each trace
call starts on a new line. Add '\n' in case the caller forgot.
Note that this explicitly limits trace output to line-by-line, it is no
longer possible to trace-print just part of a line. Until now, this was
just an implicit assumption (trace-printing part of a line worked, but
messed up the trace file if multiple threads or processes were involved).
Thread-safety / inter-process-safety is also the reason why we need to do
the prefixing and suffixing in memory rather than issuing multiple write()
calls. Write_or_whine_pipe() / xwrite() is atomic unless the size exceeds
MAX_IO_SIZE (8MB, see wrapper.c). In case of trace_strbuf, this costs an
additional string copy (which should be irrelevant for performance in light
of actual file IO).
While we're at it, rename trace_strbuf's 'buf' argument, which suggests
that the function is modifying the buffer. Trace_strbuf() currently is the
only trace API that can print arbitrary binary data (without barfing on
'%' or stopping at '\0'), so 'data' seems more appropriate.
Signed-off-by: Karsten Blees <blees@dcon.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This changes GIT_TRACE_PACK_ACCESS functionality as follows:
* supports the same options as GIT_TRACE (e.g. printing to stderr)
* no longer supports relative paths
* appends to the trace file rather than overwriting
Signed-off-by: Karsten Blees <blees@dcon.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Separate GIT_TRACE description into what it prints and how to configure
where trace output is printed to. Change other GIT_TRACE_* descriptions to
refer to GIT_TRACE.
Add descriptions for GIT_TRACE_SETUP and GIT_TRACE_SHALLOW.
Signed-off-by: Karsten Blees <blees@dcon.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The trace API currently rechecks the environment variable and reopens the
trace file on every API call. This has the ugly side effect that errors
(e.g. file cannot be opened, or the user specified a relative path) are
also reported on every call. Performance can be improved by about factor
three by remembering the environment state and keeping the file open.
Replace the 'const char *key' parameter in the API with a pointer to a
'struct trace_key' that bundles the environment variable name with
additional, trace-internal state. Change the call sites of these APIs to
use a static 'struct trace_key' instead of a string constant.
In trace.c::get_trace_fd(), save and reuse the file descriptor in 'struct
trace_key'.
Add a 'trace_disable()' API, so that packet_trace() can cleanly disable
tracing when it encounters packed data (instead of using unsetenv()).
Signed-off-by: Karsten Blees <blees@dcon.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We generally want to avoid lookup_unknown_object, because it
results in allocating more memory for the object than may be
strictly necessary.
In this case, it is used to check whether we have an
already-parsed object before calling parse_object, to save
us from reading the object from disk. Using lookup_object
would be fine for that purpose, but we can take it a step
further. Since this code was written, parse_object already
learned the "check lookup_object" optimization, so we can
simply call parse_object directly.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The point of the "index" field of struct commit is that
every allocated commit would have one. It is supposed to be
an invariant that whenever object->type is set to
OBJ_COMMIT, we have a unique index.
Commit 969eba6 (commit: push commit_index update into
alloc_commit_node, 2014-06-10) covered this case for
newly-allocated commits. However, we may also allocate an
"unknown" object via lookup_unknown_object, and only later
convert it to a commit. We must make sure that we set the
commit index when we switch the type field.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We keep a static counter to set the commit index on newly
allocated objects. However, since we also need to set the
index on any_objects which are converted to commits, let's
make the counter available as a public function.
While we're moving it, let's make sure the counter is
allocated as an unsigned integer to match the index field in
"struct commit".
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When we call lookup_commit, lookup_tree, etc, the logic goes
something like:
1. Look for an existing object struct. If we don't have
one, allocate and return a new one.
2. Double check that any object we have is the expected
type (and complain and return NULL otherwise).
3. Convert an object with type OBJ_NONE (from a prior
call to lookup_unknown_object) to the expected type.
We can encapsulate steps 2 and 3 in a helper function which
checks whether we have the expected object type, converts
OBJ_NONE as appropriate, and returns the object.
Not only does this shorten the code, but it also provides
one central location for converting OBJ_NONE objects into
objects of other types. Future patches will use that to
enforce type-specific invariants.
Since this is a refactoring, we would want it to behave
exactly as the current code. It takes a little reasoning to
see that this is the case:
- for lookup_{commit,tree,etc} functions, we are just
pulling steps 2 and 3 into a function that does the same
thing.
- for the call in peel_object, we currently only do step 3
(but we want to consolidate it with the others, as
mentioned above). However, step 2 is a noop here, as the
surrounding conditional makes sure we have OBJ_NONE
(which we want to keep to avoid an extraneous call to
sha1_object_info).
- for the call in lookup_commit_reference_gently, we are
currently doing step 2 but not step 3. However, step 3
is a noop here. The object we got will have just come
from deref_tag, which must have figured out the type for
each object in order to know when to stop peeling.
Therefore the type will never be OBJ_NONE.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The only way that "obj" can be non-NULL is if it came from
one of the lookup_* functions. These functions always ensure
that the object has the expected type (and return NULL
otherwise), so there is no need for us to set the type.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The "struct object" type implements basic object
polymorphism. Individual instances are allocated as
concrete types (or as a union type that can store any
object), and a "struct object *" can be cast into its real
type after examining its "type" enum. This means it is
dangerous to have a type field that does not match the
allocation (e.g., setting the type field of a "struct blob"
to "OBJ_COMMIT" would mean that a reader might read past the
allocated memory).
In most of the current code this is not a problem; the first
thing we do after allocating an object is usually to set its
type field by passing it to create_object. However, the
virtual commits we create in merge-recursive.c do not ever
get their type set. This does not seem to have caused
problems in practice, though (presumably because we always
pass around a "struct commit" pointer and never even look at
the type).
We can fix this oversight and also make it harder for future
code to get it wrong by setting the type directly in the
object allocation functions.
This will also make it easier to fix problems with commit
index allocation, as we know that any object allocated by
alloc_commit_node will meet the invariant that an object
with an OBJ_COMMIT type field will have a unique index
number.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Because the allocator functions for tree, blobs, etc are all
very similar, we originally used a macro to avoid repeating
ourselves. Since the prior commit, though, the heavy lifting
is done by an inline helper function. The macro does still
save us a few lines, but at some readability cost. It
obfuscates the function definitions (and makes them hard to
find via grep).
Much worse, though, is the fact that it isn't used
consistently for all allocators. Somebody coming later may
be tempted to modify DEFINE_ALLOCATOR, but they would miss
alloc_commit_node, which is treated specially.
Let's just drop the macro and write everything out
explicitly.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In order to encapsulate the setting of the unique commit index, commit
969eba63 ("commit: push commit_index update into alloc_commit_node",
10-06-2014) introduced a (logically private) intermediary allocator
function. However, this function (alloc_raw_commit_node()) was declared
as a public function, which undermines its entire purpose.
Introduce an inline function, alloc_node(), which implements the main
logic of the allocator used by DEFINE_ALLOCATOR, and redefine the macro
in terms of the new function. In addition, use the new function in the
implementation of the alloc_commit_node() allocator, rather than the
intermediary allocator, which can now be removed.
Noticed by sparse ("symbol 'alloc_raw_commit_node' was not declared.
Should it be static?").
Signed-off-by: Ramsay Jones <ramsay@ramsay1.demon.co.uk>
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Fixes a small bug affecting push to remotes which use some sort of
multi-pass authentication. In particular the bug affected SabreDAV as
configured by Box.com [1].
It must be a weird server configuration for the bug to have survived
this long. Someone should write a test for it.
[1] http://marc.info/?l=git&m=140460482604482
Signed-off-by: Abbaad Haider <abbaad@gmail.com>
Reviewed-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Rename cache_def_free to cache_def_clear as it doesn't free the struct
cache_def, but just clears its content.
Signed-off-by: Karsten Blees <blees@dcon.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The --sort tests should use the better format for >expect to maintain
indenting and ensure that no substitution is occurring. This makes
parsing and understanding the tests a bit easier.
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
A wrong '}' made our code record the results of mergetag signature
verification incorrectly.
Fix it.
Signed-off-by: Michael J Gruber <git@drmicha.warpmail.net>
Helped-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
fsck_commit_buffer() checks that the number of items in the parents
list of a commit matches the number of parent lines in its buffer or --
if a graft is used -- the number of parents in that graft. Simplify
the code by using commit_list_count() instead of counting by hand.
Also use different variables for the number of lines and the number of
list items, making it easier to compare them.
Signed-off-by: Rene Scharfe <l.s.r@web.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Build the commit_list of parents by calling commit_list_append() twice
instead of allocating and linking the items by hand. This makes the
code shorter and simpler. Rename the commit_list from parent to parents
(plural) while at it because there are two of them.
Signed-off-by: Rene Scharfe <l.s.r@web.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Clarify error message puntuation to reduce review workload.
Signed-off-by: Philip Oakley <philipoakley@iee.org>
Helped-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Only the two targets "test-lint-duplicates" and "test-lint-executable" are
currently executed when running the test target. This was done on purpose
when the TEST_LINT variable was added in 81127d74 to avoid twisted shell
scripting by developers only to avoid false positives that might result
from the rather simple minded tests, e.g. test-lint-shell-syntax. But it
looks like it might be better to include all lint tests to help developers
to detect non portable shell constructs before the patch is sent to the
list and reviewed there.
Change the TEST_LINT variable to run all lint test unless the TEST_LINT
variable is overridden. If we hit false positives more often than helping
developers to avoid non-portable code (or add less accurate or slow tests
later) we could still fall back to exclude them like 81127d74 proposed.
Signed-off-by: Jens Lehmann <Jens.Lehmann@web.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Currently only the "t[0-9][0-9][0-9][0-9]-*.sh" scripts are tested for
shell incompatibilities using the check-non-portable-shell.pl script. This
makes it easy to miss non-POSIX constructs added to one of the t/*lib*.sh
helper scripts, as they aren't automatically detected.
Fix that by adding a THELPERS variable containing all shell scripts that
aren't tests and add these to the "test-lint-shell-syntax" target too.
Signed-off-by: Jens Lehmann <Jens.Lehmann@web.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Fixes to a topic that is already in 'master'.
* dt/refs-check-refname-component-sse-fix:
refs: fix valgrind suppression file
refs.c: handle REFNAME_REFSPEC_PATTERN at end of page
Add 'verify-commit' to be used in a way similar to 'verify-tag' is
used. Further work on verifying the mergetags might be needed.
* mg/verify-commit:
t7510: test verify-commit
t7510: exit for loop with test result
verify-commit: scriptable commit signature verification
gpg-interface: provide access to the payload
gpg-interface: provide clear helper for struct signature_check
"git clone -b brefs/tags/bar" would have mistakenly thought we were
following a single tag, even though it was a name of the branch,
because it incorrectly used strstr().
* jc/fix-clone-single-starting-at-a-tag:
builtin/clone.c: detect a clone starting at a tag correctly
* jk/repack-pack-keep-objects:
repack: s/write_bitmap/&s/ in code
repack: respect pack.writebitmaps
repack: do not accidentally pack kept objects by default
We can make the parsing of the --sort parameter a bit more
readable by having skip_prefix keep our pointer up to date.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When we encounter an error in remote-curl, we generally just
report it to stderr. There is no need for the user to care
that the "could not connect to server" error was generated
by git-remote-https rather than a function in the parent
git-fetch process.
However, when the error is in the protocol between git and
the helper, it makes sense to clearly identify which side is
complaining. These cases shouldn't ever happen, but when
they do, we can make them less confusing by being more
verbose.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We usually prefix our error messages with "error: ", but
many error messages from remote-curl are simply printed with
fprintf. This can make the output a little harder to read
(especially because such message may be intermingled with
errors from the parent git process).
There is no reason to avoid error(), as we are already
calling it many places (in addition to libgit.a functions
which use it).
While we're adjusting the messages, we can also drop the
capitalization which makes them unlike other git error
messages.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The parent git process is supposed to send us an empty line
to indicate that the conversation is over. However, the
parent process may die() if there is a problem with the
operation (e.g., we try to fetch a ref that does not exist).
In this case, it produces a useful message, but then
remote-curl _also_ produces an unhelpful message:
$ git pull origin matser
fatal: couldn't find remote ref matser
Unexpected end of command stream
The "right" way to fix this is to teach the parent git to
always cleanly close the connection to the helper, letting
it know that we are done. Implementing that is rather
clunky, though, as it would involve either replacing die()
operations with returning errors up the stack (until we
disconnect the transport), or adding an atexit handler to
clean up any transport helpers left open.
It's much simpler to just suppress the EOF message in
remote-curl. It was not added to address any real-world
situation in the first place, but rather a "we should
probably report unexpected things" suggestion[1].
It is the parent git which drives the operation, and whose
exit value actually matters. If the parent dies, then the
helper has no need to complain (except as a debugging aid).
In the off chance that the pipe is closed without the parent
dying, it can still notice the non-zero exit code.
[1] http://article.gmane.org/gmane.comp.version-control.git/176036
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
