[Bug gcov-profile/89961] When "--intermediate-format" is used "--preserve-paths"/"--hash-filenames" is ignored

[Bug gcov-profile/89961] When "--intermediate-format" is used "--preserve-paths"/"--hash-filenames" is ignored

[xaizek at posteo dot net]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89961

xaizek <xaizek at posteo dot net> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |xaizek at posteo dot net

--- Comment #11 from xaizek <xaizek at posteo dot net> ---
> Closing this as fixed.

This isn't fixed. I don't think the problem was actually understood.

Say you have:

 * src/myclass.o
 * tests/myclass.o

If you run `gcov -p -x --json-format src/myclass.o` you get
myclass.gcov.json.gz for that file, running `gcov -p -x --json-format
tests/myclass.o` gives you myclass.gcov.json.gz for the test. Running `gcov -p
-x --json-format src/myclass.o tests/myclass.o` gives myclass.gcov.json.gz *FOR
THE TEST ONLY*, because gcov created output for src/myclass.o and then replaced
it.

>From reading your comment I thought that maybe gcov combines information from
two files before writing the result to address this issue, but it doesn't.
Results you get depends on the order in which files are specified.

Either -p and -x need to be  obeyed or reports need to be merged. If combining
is implemented, might as well combine all reports and write a single file
specified on command line (--json-format=myfile.gz or --json-format=- (no
compression needed)).

[Bug gcov-profile/89961] When "--intermediate-format" is used "--preserve-paths"/"--hash-filenames" is ignored

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89961

Martin Liška <marxin at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|FIXED                       |---
             Status|RESOLVED                    |REOPENED

--- Comment #12 from Martin Liška <marxin at gcc dot gnu.org> ---
I'll take a look at it.

[Bug gcov-profile/89961] When "--intermediate-format" is used "--preserve-paths"/"--hash-filenames" is ignored

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89961

--- Comment #13 from Martin Liška <marxin at gcc dot gnu.org> ---
(In reply to xaizek from comment #11)
> > Closing this as fixed.
> 
> This isn't fixed. I don't think the problem was actually understood.
> 
> Say you have:
> 
>  * src/myclass.o
>  * tests/myclass.o
> 

Looking into the documentation:
https://gcc.gnu.org/onlinedocs/gcc/Invoking-Gcov.html#Invoking-Gcov

```
gcov should be run with the current directory the same as that when you invoked
the compiler. Otherwise it will not be able to locate the source files. gcov
produces files called mangledname.gcov in the current directory.
```

> If you run `gcov -p -x --json-format src/myclass.o` you get
> myclass.gcov.json.gz for that file, running `gcov -p -x --json-format
> tests/myclass.o` gives you myclass.gcov.json.gz for the test. Running `gcov
> -p -x --json-format src/myclass.o tests/myclass.o` gives
> myclass.gcov.json.gz *FOR THE TEST ONLY*, because gcov created output for
> src/myclass.o and then replaced it.

This example violates that because you would end up with clasing myclass.o.

> 
> From reading your comment I thought that maybe gcov combines information
> from two files before writing the result to address this issue, but it
> doesn't. Results you get depends on the order in which files are specified.
> 
> Either -p and -x need to be  obeyed or reports need to be merged. If
> combining is implemented, might as well combine all reports and write a
> single file specified on command line (--json-format=myfile.gz or
> --json-format=- (no compression needed)).

What you can use is -t argument that will output to stdout where you'll get 2
JSON objects.

[Bug gcov-profile/89961] When "--intermediate-format" is used "--preserve-paths"/"--hash-filenames" is ignored

[xaizek at posteo dot net]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89961

--- Comment #14 from xaizek <xaizek at posteo dot net> ---
> This example violates that because you would end up with clasing myclass.o.

I don't think anything is violated, because GCC has `-o` option. It's
equivalent in GCOV is mangling:

> The mangledname part of the output file name is usually simply the source file name, but can be something more complicated if the ‘-l’ or ‘-p’ options are given.

Neither `-p` nor `-l` works for JSON output of GCOV, which causes loss of
output files. I fail to see has this can not be a bug.

> What you can use is -t argument that will output to stdout where you'll get 2 JSON objects.

Didn't know it was added, thanks. When it's available, it's worth using.

[Bug gcov-profile/89961] When "--intermediate-format" is used "--preserve-paths"/"--hash-filenames" is ignored

[xaizek at posteo dot net]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89961

--- Comment #15 from xaizek <xaizek at posteo dot net> ---
> I don't think anything is violated, because GCC has `-o` option.

Actually, I tried it and GCC can produce such weird results as well, but
documentation talks about file discovery, not about reproducing this behaviour.

[Bug gcov-profile/89961] When "--intermediate-format" is used "--preserve-paths"/"--hash-filenames" is ignored

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89961

--- Comment #16 from Martin Liška <marxin at gcc dot gnu.org> ---
All right, so what's different about JSON and non-JSON mode:

1) normal mode:

gcov src/a.c test/a.c

First, gcda/gcno files are searched for both the arguments (so src/a.gcda, ..)
and then output is emitted for all mentioned source files (can be .c or .h
files). Note the output files x.gcov are based on *SOURCE* file names, so e.g.:

File 'tramp3d-v4.cpp'
Lines executed:0.00% of 7339
Creating 'tramp3d-v4.cpp.gcov'

File '/usr/include/c++/6/ext/new_allocator.h'
Lines executed:0.00% of 14
Creating 'new_allocator.h.gcov'

File '/usr/include/c++/6/ext/aligned_buffer.h'
Lines executed:0.00% of 8
Creating 'aligned_buffer.h.gcov'

File '/usr/include/c++/6/bits/move.h'
Lines executed:0.00% of 11
Creating 'move.h.gcov'
...


2) in JSON mode, we basically emit JSON output for each GCDA file (deduced from
arguments). What's currently wrong is that for
$ gcov src/a.xxxxx b.yyyyyy --json-format

The following 2 files should be created (right now b.gcov.json.gz and
a.gcov.json.gz are created)

b.gcov.json.gz
src/a.gcov.json.gz

and so -p and -x does not play any role here.

Does it make sense to you? If so, I'm prepared sending a patch candidate.

[Bug gcov-profile/89961] When "--intermediate-format" is used "--preserve-paths"/"--hash-filenames" is ignored

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89961

--- Comment #17 from Martin Liška <marxin at gcc dot gnu.org> ---
> $ gcov src/a.xxxxx b.yyyyyy --json-format

this ought to be src/a.c b.c (but as you noticed gcov deduces .gcda/.gcno files
by skipping the last dot component of a path). That's why it works as well.

[Bug gcov-profile/89961] When "--intermediate-format" is used "--preserve-paths"/"--hash-filenames" is ignored

[xaizek at posteo dot net]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89961

--- Comment #18 from xaizek <xaizek at posteo dot net> ---
> The following 2 files should be created (right now b.gcov.json.gz and
> a.gcov.json.gz are created)
> 
> b.gcov.json.gz
> src/a.gcov.json.gz
> 
> and so -p and -x does not play any role here.

Now you'll need to search for output recursively, but this will fix the issue,
although I'm not sure what will happen if you run `gcov` outside of build tree
(I remember the docs, but isn't intermediate format is more usable because it
doesn't need sources and so the restriction on where to run `gcov` doesn't
apply?).

I still don't understand why not just make `-p` and `-x` work?  The options
are there anyway.

By the way, I worked around the issue in May by not invoking `gcov` with two
files that are named identically.  It works even with GCC 8 for me (although
not
in one call), but people will continue to use `-p` and `-x` and get surprising
results.

[Bug gcov-profile/89961] When "--intermediate-format" is used "--preserve-paths"/"--hash-filenames" is ignored

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89961

--- Comment #19 from Martin Liška <marxin at gcc dot gnu.org> ---
Created attachment 51316
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=51316&action=edit
Patch candidate

All right, I've created a patch that implements -x and -l options for JSON
format.
@xaizek: Can you please test it?

[Bug gcov-profile/89961] When "--intermediate-format" is used "--preserve-paths"/"--hash-filenames" is ignored

[xaizek at posteo dot net]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89961

--- Comment #20 from xaizek <xaizek at posteo dot net> ---
The patch works! I think you actually implemented behaviour of `-p` for `-l`
(you hash path in the code and don't process includes). `-l` doesn't seem to
make much sense for JSON, because all includes will be in the same object
anyway, while `-p` does.

Thank you for making it work. I really think that's what users of gcov expect
from those options.

By the way, since this is another change related to automatic invocation of
gcov. Just as lcov does it, I recently started to parse output of `gcov --help`
with regexes in uncov to detect JSON support. This works, but it would be more
reliable if gcov could print its capabilities in some fixed form to don't make
tools do such things. Of course existing parsing code won't go away, but it
might become a fallback for a more robust path.

[Bug gcov-profile/89961] When "--intermediate-format" is used "--preserve-paths"/"--hash-filenames" is ignored

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89961

--- Comment #21 from Martin Liška <marxin at gcc dot gnu.org> ---
(In reply to xaizek from comment #20)
> The patch works! I think you actually implemented behaviour of `-p` for `-l`
> (you hash path in the code and don't process includes). `-l` doesn't seem to
> make much sense for JSON, because all includes will be in the same object
> anyway, while `-p` does.

Oh, you are right, it should be fixed in the updated version of the patch.
Moreover, I also made a refactoring for non-JSON code where filenames are
created.

> 
> Thank you for making it work. I really think that's what users of gcov
> expect from those options.

Sure, you are right.

> 
> By the way, since this is another change related to automatic invocation of
> gcov. Just as lcov does it, I recently started to parse output of `gcov
> --help` with regexes in uncov to detect JSON support. This works, but it
> would be more reliable if gcov could print its capabilities in some fixed
> form to don't make tools do such things. Of course existing parsing code
> won't go away, but it might become a fallback for a more robust path.

Right now, one can easily detect JSON format with `gcov --help | grep
json-format`, am I right? What type of capability would you appreciate?

[Bug gcov-profile/89961] When "--intermediate-format" is used "--preserve-paths"/"--hash-filenames" is ignored

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89961

Martin Liška <marxin at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #51316|0                           |1
        is obsolete|                            |

--- Comment #22 from Martin Liška <marxin at gcc dot gnu.org> ---
Created attachment 51323
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=51323&action=edit
Patch candidate v2

[Bug gcov-profile/89961] When "--intermediate-format" is used "--preserve-paths"/"--hash-filenames" is ignored

[xaizek at posteo dot net]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89961

--- Comment #23 from xaizek <xaizek at posteo dot net> ---
> it should be fixed in the updated version of the patch.

Yes, it is.

> Right now, one can easily detect JSON format with
> `gcov --help | grep json-format`, am I right? What type of capability would
> you appreciate?

Correct, but `--help` and `--version` are formatted for humans, not machines.
This can lead to false positives or negatives.  It forces every application
that
calls gcov and wants to support wide range of versions to embed its copy of
knowledge about gcov's development history.  And `--help` won't tell if `-x`
works with `--json-format`, that's why `--version` needs to be used as well.

For consistency, since there is output for machines, it would be nice to have a
way to know how to get it and what to expect from it for machines as well.

I'd include any addition/removal/change between versions that makes difference,
like "has-json-format-option", "can-hash-json-files", "has-stdout-option".
Basically a hard-coded list that's easy to check.

This isn't so much for existing releases (unless it's backported), but in
anticipation of future changes that can affect the caller.  I don't have an
example of such interface in apps, just wanted to bring it up for your
consideration.

[Bug gcov-profile/89961] When "--intermediate-format" is used "--preserve-paths"/"--hash-filenames" is ignored

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89961

--- Comment #24 from Martin Liška <marxin at gcc dot gnu.org> ---
> Correct, but `--help` and `--version` are formatted for humans, not machines.
> This can lead to false positives or negatives.  It forces every application
> that
> calls gcov and wants to support wide range of versions to embed its copy of
> knowledge about gcov's development history.  And `--help` won't tell if `-x`
> works with `--json-format`, that's why `--version` needs to be used as well.

I see your point, but I would rather recommend doing a "configure check", I
mean having
a bash snippet that tests various features.

[Bug gcov-profile/89961] When "--intermediate-format" is used "--preserve-paths"/"--hash-filenames" is ignored

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89961

--- Comment #25 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Martin Liska <marxin@gcc.gnu.org>:

https://gcc.gnu.org/g:b777f228b481ae881a7fbb09de367a053740932c

commit r12-3038-gb777f228b481ae881a7fbb09de367a053740932c
Author: Martin Liska <mliska@suse.cz>
Date:   Tue Aug 17 16:24:26 2021 +0200

    gcov: fix output location for JSON mode.

            PR gcov-profile/89961

    gcc/ChangeLog:

            * gcov.c (make_gcov_file_name): Rewrite using std::string.
            (mangle_name): Simplify, do not used the second argument.
            (strip_extention): New function.
            (get_md5sum): Likewise.
            (get_gcov_intermediate_filename): Handle properly -p and -x
            options.
            (output_gcov_file): Use string type.
            (generate_results): Likewise.
            (md5sum_to_hex): Remove.

[Bug gcov-profile/89961] When "--intermediate-format" is used "--preserve-paths"/"--hash-filenames" is ignored

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89961

Martin Liška <marxin at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
      Known to fail|                            |11.2.0
      Known to work|                            |12.0

--- Comment #26 from Martin Liška <marxin at gcc dot gnu.org> ---
Fixed on master so far, I'm planning to do a backport to gcc-11 branch.

[Bug gcov-profile/89961] When "--intermediate-format" is used "--preserve-paths"/"--hash-filenames" is ignored

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89961

--- Comment #27 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-11 branch has been updated by Martin Liska
<marxin@gcc.gnu.org>:

https://gcc.gnu.org/g:f7134a12cd633e8ffdcee1af7e294e9ab60647c0

commit r11-8892-gf7134a12cd633e8ffdcee1af7e294e9ab60647c0
Author: Martin Liska <mliska@suse.cz>
Date:   Tue Aug 17 16:24:26 2021 +0200

    gcov: fix output location for JSON mode.

            PR gcov-profile/89961

    gcc/ChangeLog:

            * gcov.c (make_gcov_file_name): Rewrite using std::string.
            (mangle_name): Simplify, do not used the second argument.
            (strip_extention): New function.
            (get_md5sum): Likewise.
            (get_gcov_intermediate_filename): Handle properly -p and -x
            options.
            (output_gcov_file): Use string type.
            (generate_results): Likewise.
            (md5sum_to_hex): Remove.

    (cherry picked from commit b777f228b481ae881a7fbb09de367a053740932c)

[Bug gcov-profile/89961] When "--intermediate-format" is used "--preserve-paths"/"--hash-filenames" is ignored

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89961

Martin Liška <marxin at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|---                         |FIXED
      Known to fail|11.2.0                      |
      Known to work|                            |11.2.1
             Status|REOPENED                    |RESOLVED

--- Comment #28 from Martin Liška <marxin at gcc dot gnu.org> ---
Fixed on gcc-11 branch, not planning backporting more.