issue with run-tests: failures in older linux distros?

[wdlkmpx]

seeing the build logs of several pull requests, the only part that seems to make a difference is zlib but the compiler might be configured in a different way

• slackware current (2020-06) with zlib 1.2.11

# ./rebuild.sh
# ./run-tests.sh 
Running test ./test/v0/the-feeble-files-spanish.sh...succeeded
Running test ./test/v0/baldurs_gate_patch_v1_1_4315_international.sh...succeeded
Running test ./test/v0/avigomanager.sh...succeeded
Running test ./test/v0/wireplay.sh...succeeded
Running test ./test/v5/CVE-2015-1386/CVE-2015-1386.sh...succeeded

• debian stretch (2017) with zlib 1.2.8

  # ./rebuild.sh
  # ./run-tests.sh 
  Running test ./test/v0/the-feeble-files-spanish.sh...FAILED with code 2
  Running test ./test/v0/baldurs_gate_patch_v1_1_4315_international.sh...succeeded
  Running test ./test/v0/avigomanager.sh...FAILED with code 3
  Running test ./test/v0/wireplay.sh...FAILED with code 3
  Running test ./test/v5/CVE-2015-1386/CVE-2015-1386.sh...succeeded

[wdlkmpx]

The commands were ran in a clean source tree (git clean -dfx), using the master branch, with a shared library linked to openssl and zlib.

I don't understand why debian stretch is using 1.2.8... I think 1.2.11 was already available back then https://packages.debian.org/stretch/zlib1g

The evidence potentially shows that only the latest stable zlib version should be used, perhaps it should be integrated into unshield for consistency

[twogood]

That is very interesting.

[wdlkmpx]

It's definitely zlib, I locally reimplemented autotools support. I compiled a static zlib 1.2.11 that is linked to a static libunshield

• debian stretch (2017) with static zlib 1.2.11

# ./rebuild.sh autotools
# ./run-tests.sh 
Running test ./test/v0/the-feeble-files-spanish.sh...succeeded
Running test ./test/v0/baldurs_gate_patch_v1_1_4315_international.sh...succeeded
Running test ./test/v0/avigomanager.sh...succeeded
Running test ./test/v0/wireplay.sh...succeeded
Running test ./test/v5/CVE-2015-1386/CVE-2015-1386.sh...succeeded

modified rebuild.sh

if test -z "$1" ; then
    export CFLAGS="-Wall -Werror -ggdb3"
    mkdir -p build
    cd build
    cmake -DCMAKE_INSTALL_PREFIX:PATH=/var/tmp/unshield .. && make && make install
else
    # param = autotools
    if test -f zlib/libz.a ; then
        # zlib/libz.a
        # zlib/conf.h
        # zlib/zlib.h
        export ZLIB_LIBS="-L$(pwd)/zlib -lz"
        export ZLIB_CFLAGS="-L$(pwd)/zlib"
    fi
    if ! test -f configure ; then
        ./autogen.sh
    fi
    ./configure --prefix=/var/tmp/unshield --enable-static --disable-shared && \
    make clean && \
    make install
fi

[wdlkmpx]

No, something was wrong. If there is a system libunshield, unshield will load /usr/lib/liblibunshield.so instead of /var/tmp/unshield/lib/liblibunshield.so, this makes some tests fail.

That's why rebuild.sh must create a static libunshield: cmake -DBUILD_STATIC=ON ....

[twogood]

Ahh damn :)

[wdlkmpx]

How to fix this:

rebuild.sh must create a static libunshield: cmake -DBUILD_STATIC=ON ....

This and other random changes I happen to make will be in my repo, it takes too much time to test, explain and 'sell' this stuff. I'm accumulating too many changes, which will require too many branches and PR's, and this will confuse me a lot. stuff can be cherry-picked afterwards.