libgit2 @1.5.0: install_name changed, dependents need rev bump

[tventers]

$ port -v
MacPorts 2.7.2

$ sudo port clean all
Password:

$ sudo port install rav1e
Password:

--->  Computing dependencies for rav1e
--->  Fetching distfiles for rav1e
--->  Verifying checksums for rav1e
--->  Extracting rav1e
--->  Configuring rav1e
--->  Building rav1e
Error: Failed to build rav1e: command execution failed   
Error: See /opt/local/var/macports/logs/_opt_local_var_macports_sources_github.com_macports_macports-ports_multimedia_rav1e/rav1e/main.log for details.
Error: Follow https://guide.macports.org/#project.tickets if you believe there
is a bug.
Error: Processing of port rav1e failed

[tventers]

Install logfile of rav1e

[jmroot (Joshua Root)]

Looks like libgit2 broke binary compatibility (previous install_name /opt/local/lib/libgit2.1.3.dylib, current one /opt/local/lib/libgit2.1.5.dylib.) So all the dependents (such as cargo) need to be rev bumped.

[mascguy (Christopher Nielsen)]

The broken binary in question - cargo-cbuild - is provided by port cargo-c. And AFAIK, other ports like rust and cargo are fine. (If they weren't, builds would be failing everywhere, which they aren't.)

Also confirmed that the buildbot binaries for cargo-c are indeed stale, and need to be rebuilt. So I'll rev-bump that port.

[Christopher Nielsen <mascguy@…>]

In df493c90ab7c8a2c401b4efdda437fc3b7fb71e2/macports-ports (master):

cargo-c: rev-bump for libgit2 ABI change
Fixes: #65733

[jmroot (Joshua Root)]

What about the other 20+ dependents?

[jmroot (Joshua Root)]

dura for one is still broken. All the libgit2 dependents that haven't been updated since the last libgit2 update on August 1st need to be rev bumped (or have their dependencies fixed if they don't actually link with it).

[mascguy (Christopher Nielsen)]

Replying to jmroot:

dura for one is still broken. All the libgit2 dependents that haven't been updated since the last libgit2 update on August 1st need to be rev bumped (or have their dependencies fixed if they don't actually link with it).

I've already spent multiple hours on this - including queuing rebuilds for 10.7, etc - and need to move on. Have at it.

[jmroot (Joshua Root)]

The implication was not that it's your job, but the problem does still exist.

[mascguy (Christopher Nielsen)]

Replying to jmroot:

The implication was not that it's your job, but the problem does still exist.

Understood. I simply wanted to let folks know that I can't spend more time on it, as much as I'd like to!

[Christopher Nielsen <mascguy@…>]

In 830e1ffafc404c44c766ff5cc6814a3d73036577/macports-ports (master):

libgit2 dependents: rev-bump for ABI change
See: #65733

[mascguy (Christopher Nielsen)]

Since the rev-bumps are a no-brainer, I took care of that.

As for validating deps for these ports, presumably most of them do link with libgit2, given their purposes. So perhaps we're mostly good on that front already.

However, I'm curious as to why cargo and rust weren't broken: A quick non-exhaustive check of related binaries with otool -L, didn't show a dylib reference. So are these components dynamically loading libgit2*.dylib at runtime, or did I simply not check enough of them...?

[mascguy (Christopher Nielsen)]

Replying to mascguy:

However, I'm curious as to why cargo and rust weren't broken: A quick non-exhaustive check of related binaries with otool -L, didn't show a dylib reference. So are these components dynamically loading libgit2*.dylib at runtime, or did I simply not check enough of them...?

Hmmm, I'm seeing some unexpected build failures for Rust-based ports, which suggest that perhaps cargo and/or rust do need to be rebuilt. (While they don't appear to be broken in-general, the rustlib utilizing libgit2 does appear that it is.)

So, I'll rev-bump rust and cargo, and then queue rebuilds for the affected ports. (Was hoping to avoid that, as rust ties up our buildbots for hours on end. Sigh...)

[Christopher Nielsen <mascguy@…>]

In b27d880d38e6b06bbe025cb44ee0d0e4feaa2f6a/macports-ports (master):

rust/cargo: rev-bump for libgit2 ABI change
See: #65733

[mascguy (Christopher Nielsen)]

Replying to mascguy:

So, I'll rev-bump rust and cargo, and then queue rebuilds for the affected ports. (Was hoping to avoid that, as rust ties up our buildbots for hours on end. Sigh...)

Queued rebuilds across-the-board, for the three ports affected: exa, gitui, and mergestat.

[mascguy (Christopher Nielsen)]

Assigning back to myself, since I'm spearheading the effort.

[mascguy (Christopher Nielsen)]

Of note, gitui is broken across-the-board; tracked separately by: issue:65654

[mascguy (Christopher Nielsen)]

Port mergestat is also broken across-the-board; tracked by new ticket: issue:65756

[Gandoon (Erik Hedlund)]

Just for information, I noticed this a few weeks ago and had to revert to libgit2 @1.3.0_1 as ibgit2 @1.5.0_0 broke exa @0.10.1_2+doc+git (turned out it was the +git that messed things up).

I was planning on trying to see if I could figure out if I could find a workaround, but I never had time for it, so I just reverted for the time being and never submitted a ticket. Good that someone got around to it. I will keep an eye out for fixes.

[mascguy (Christopher Nielsen)]

Rust gurus, do we also need to update the various git2-related crates, now that libgit2 has been updated to 1.5.x? Or will everything "just work," with the versions we're currently using?

[mascguy (Christopher Nielsen)]

Replying to mascguy:

Rust gurus, do we also need to update the various git2-related crates, now that libgit2 has been updated to 1.5.x? Or will everything "just work," with the versions we're currently using?

Everything looks good now, after all rev-bumps and rebuilds. Closing as fixed.