Wireshark-dev: Re: [Wireshark-dev] Memory leak debugging - current master passes all tests!
From: Peter Wu <[email protected]>
Date: Wed, 30 Jan 2019 12:19:27 +0100
On Tue, Jan 29, 2019 at 08:59:30AM +0100, Dario Lombardo wrote:
> > This does not mean that there are no more memory leaks though (we have
> > pretty bad test coverage)
> Have you ever tried to compute the coverage with, eg., gcov/lcov? I'm
> working on it but it's still a WIP.
> There are also public platforms (as coveralls) that help keeping track of
> it over time/over history.

I have used a coverage checker (llvm-cov) during development of a single
dissector, but not the whole project. From my notes:

    # epan/dissectors/CMakeList.txt after "set_target_properties(dissectors PROPERTIES"
    set(COV_FLAGS "-fprofile-instr-generate -fcoverage-mapping")
    set_source_files_properties(packet-wireguard.c PROPERTIES COMPILE_FLAGS "${WERROR_COMMON_FLAGS} ${COV_FLAGS}")

    # epan/CMakeLists.txt after "set_target_properties(epan PROPERTIES"
    set_target_properties(epan PROPERTIES LINK_FLAGS "${WS_LINK_FLAGS} ${COV_FLAGS}")

    # run program and check results with
    llvm-profdata merge -sparse default.profraw -o default.profdata
    llvm-cov show \
        /tmp/wsbuild/epan/dissectors/CMakeFiles/dissectors.dir/packet-wireguard.c.o \
        -instr-profile=default.profdata -use-color=1 \
        /tmp/wireshark/epan/dissectors/packet-wireguard.c | less -R

Though even without using external tools, a good starting point would be
adding basic tests for everything that uses a UAT. Stig recently fixed
some memleaks in packet-lwm2mtlv.c which were not caught by tests...
because there were none! ;)

On Tue, Jan 29, 2019 at 09:15:50AM -0500, Jeff Morriss wrote:
> Nice info.  I had wondered at some points how to run ASAN but not enough to
> do the (probably minimal) research.  Could I suggest putting the info in a
> Wiki page?  (When I look for info I look in the READMEs and the wiki; I
> rarely search the email archives unless I remember what I was looking for
> being there.)

I think I have found an appropriate place for it here:

Feel free to make more edits as needed :)
Kind regards,
Peter Wu