lzmalib is present
AbandonedPublic
Actions

Authored by dbaio on Sep 18 2020, 9:46 PM.

Details

Reviewers

koobs

Group Reviewers

Python

Summary

lang/python3*: Fix build when archivers/lzmalib is installed

The lzma module fails to build when archivers/lzmalib is
installed:

running build_ext
[...]
*** WARNING: renaming "_lzma" since importing it failed: 
  build/lib.freebsd-12.1-RELEASE-p8-amd64-3.7/_lzma.so: 
    Undefined symbol "lzma_check_is_supported"

This change configures the lzma module to always build against base
(libarchive) via  WRKSRC/Modules/Setup.local.

Eventually we can build it against archivers/lzma, but currently
the port does not provide header and libraries.

The change does no apply to Python 2.7 as it does not provide the
lzma module

 PR:   209355, 243817
 Reported by:  Eir Nym <eirnym@gmail.com>
 Reported by:  doctor@doctor.nl2k.ab.ca
 Submitted by: loader (earlier version)
 Reviewed by:  koobs (Python)
 Differential Revision:	D26477

Test Plan

testport: OK (poudriere: 11, 12, CURRENT; i386, amd64)

Diff Detail

Repository

rP FreeBSD ports repository

Lint

Lint Skipped

Unit

Tests Skipped

Event Timeline

dbaio created this revision.Sep 18 2020, 9:46 PM

Herald added a subscriber: mat. · View Herald TranscriptSep 18 2020, 9:46 PM

dbaio requested review of this revision.Sep 18 2020, 9:46 PM

dbaio edited the test plan for this revision. (Show Details)Sep 19 2020, 11:00 AM

koobs added a reviewer: koobs.Sep 20 2020, 4:26 AM

Adding clarifications, requested from @koobs

koobs added inline comments.Sep 20 2020, 1:22 PM

lang/python36/Makefile
132–135	Since this issue and changeset is all about 'building with base', and since there's no port (at present, as archivers/lzma only provides a binary, not headers), is there any reason we shouldn't just explicitly and unconditionally point the lzma module to base?

koobs edited the summary of this revision. (Show Details)Sep 20 2020, 1:24 PM

koobs edited the summary of this revision. (Show Details)Sep 21 2020, 7:31 AM

koobs edited the test plan for this revision. (Show Details)

koobs added inline comments.

lang/python36/Makefile
132–135	We dont need the conditional if we always want to build against base (and provide the relevent -I -l args for the lzma module) regardless, no?

dbaio added inline comments.Sep 21 2020, 10:58 PM

lang/python36/Makefile
132–135	This issue occurs when not using Poudriere, on a very specific scenario, I would recommend letting this way. So it won't change anything (to the current users), and _lzma will continue to be built by setup.py respecting system flags.

koobs added inline comments.Sep 22 2020, 5:09 AM

lang/python36/Makefile
132–135	I think the explicit args for lzma in Setup.local has always the been the correct and appropriate thing to do (which is why i suggested it), and is the correct and permanent resolution. I don't think this should be conditionalized unless and until we support lzmalibs from base and ports Additionally, it would be nice to confirm that poudriere passes with a hard BUILD_DEPENDS on archivers/lzmalib with the conditional removed in this block