On Fri, 22 Sep 2023 07:45:43 GMT, Magnus Ihse Bursie <i...@openjdk.org> wrote:
>> It may be possible to read some elf header about binutils version. >> I.e. do make install from our build system, then reading out version from >> elf section and pass that into our build. >> As we need to know the signature of init_disassemble_info() which is >> different between binutils versions. > > I'm still not really happy about this. The old solution without .libs has > worked before -- has anything changed in newer versions of binutils? > > Also, we support building binutils in place as a convenience, but it should > also be possible to just set a hsdis path, and in that case we cannot presume > that the .libs layout is kept. > > I suggest you change this to both eat the cake and have it -- first check for > the lib in the original location (without .libs), and if it is not found > there, check in .libs. This is perhaps not necessary here in > LIB_BUILD_BINUTILS, but it definitely is in LIB_SETUP_HSDIS_BINUTILS. It will > make the code a few lines longer but more robust. > It may be possible to read some elf header about binutils version. I.e. do > make install from our build system, then reading out version from elf section > and pass that into our build. As we need to know the signature of > init_disassemble_info() which is different between binutils versions. Trying to understand what you are really saying here. If we accept this patch, it will no longer be possible to use a normally installed binutils, since the BFD_VERSION constant (via bfdver.h) is not available? That is not good; the idea of building binutils locally was supposed to be a convenience, not a requirement. ------------- PR Review Comment: https://git.openjdk.org/jdk/pull/15138#discussion_r1334029663