Gracefully handle outdated .ninja_log during '-t recompact'

[jheydebrand]

Fixes #2048

I ran into a similar issue when migrating existing workspaces from ninja 1.9.0 to the latest revision from Git.

This change will cause NinjaMain::OpenBuildLog() to not call build_log_.Recompact() after build_log_.Load() just deleted that file because it contained an invalid log version.

[digit-google]

Thank you. The fix looks good to me, but can you add a regression test in misc/output_test.py to verify that this works correctly?

E.g.: perform a build, then overwrite the version bytes in the build log, then run the recompact tool and verify it does not report an error?

[jheydebrand]

I added a test as suggested, @digit-google
However, I could run it only on Windows locally - hopefully test test runs fine on Linux, too.

[digit-google]

Thank you, this looks good to me. @jhasse what do you think?

[jheydebrand]

Should I try to work around Python 3.6 in AppVeyor not knowing text=True?

[jhasse]

Dropping Ubuntu 14.04, 16.04 and 18.04 in favor of 20.04, 22.04 and 24.04 is something I would want to do soon, so you might also just wait. But I think we'd also have to run these tests on RHEL 7 (or 8 soon), I don't know what Python version they are using.

[jheydebrand]

Looks like RHEL 7 supports Python 3.8 per https://access.redhat.com/documentation/en-us/red_hat_software_collections/3/html-single/3.5_release_notes/index#sect-RHSCL-Changes-python

Depending on your timeline for dropping the older distros, I'd offer to get the test to work with the older Python version to wrap this PR up.

[digit-google]

Aargh, I was not aware of this limitation, remove the text=True for now then, this is not a good reason to block this PR. Thanks.

[jhasse]

See #2399

[jhasse]

Can you rebase against / merge master?

[jheydebrand]

Looking good now :)

[digit-google]

LGTM now, thank you.