It Looks Like GCC's Long-Awaited Git Conversion Could Happen This Weekend

[CommunityMember]

It's worth noting that the work Maxim has done on his git-svn conversion has helped improve reposurgeon and the gcc git conversion in general too, which is probably the best possible outcome of all this as it will apparently improve gcc history for everyone in terms of correctness of attribution going all the way back to the pre-CVS days where there were apparently just omnibus -- and somewhat haphazard? -- code-drops with release tarballs.

For some time the code/history conversion was more or less equivalent and complete between the two leading alternatives. Attribution and references (sometimes needing to parse the changelogs or code itself, for such things like determining authors rather than just committers and extracting the bug/issue/enhancement references and matching multiple email addresses to the same person) was the long tail, with what are now some better heuristics, but more importantly a lot of domain specific adjustments (i.e. one off fixes), to the conversion process for the gcc use case. No one is so arrogant (well, there is always one) to believe there are still not some artifacts if you spent the time to look really closely, but they are of a small enough number that no one is willing to spend another month trying to address them, and then another month trying to address an even smaller number, and then another month.....

[Spooktra]

He didn't get it, so forced others to rewrite Reposurgeon to Golang.

I thought some sucker, I mean Good Samaritan, bought him a new computer.

[CommunityMember]

I thought some sucker, I mean Good Samaritan, bought him a new computer.

That did happen in the past during one of his various asks. But as I understand it, the actual (really final this time) conversion will be done on real server class hardware and not in someone's basement.

[timofonic]

That did happen in the past during one of his various asks. But as I understand it, the actual (really final this time) conversion will be done on real server class hardware and not in someone's basement.

Do you mean the real server class hardware is by some sucker company? It makes sense...

[L_A_G]

Experienced developers joined the effort. It should surprise no one that actual talent can git-r-done.

If they actually meant to get it done, then why was it delegated to ESR in the first place? Assigning projects to people who are lazy and/or incompetent is a well known strategy used by managers who wan't to kill a project, but don't want to do it outright.

[ermo]

For some time the code/history conversion was more or less equivalent and complete between the two leading alternatives.

Attribution and references (sometimes needing to parse the changelogs or code itself, for such things like determining authors rather than just committers and extracting the bug/issue/enhancement references and matching multiple email addresses to the same person) was the long tail, with what are now some better heuristics, but more importantly a lot of domain specific adjustments (i.e. one off fixes), to the conversion process for the gcc use case.

No one is so arrogant (well, there is always one) to believe there are still not some artifacts if you spent the time to look really closely, but they are of a small enough number that no one is willing to spend another month trying to address them, and then another month trying to address an even smaller number, and then another month.....

Maxim's conversion was apparently a very useful reference for tweaking the heuristics in reposurgeon by helping to flag some overly conservative lifts in reposurgeons GCC-specific fixups, thus leading to improved attributions. Whether or not this trip down the long tail of history lane was worth it is not for me to judge.

As for artifacts, well, the threshold for diminishing returns has clearly been reached. AIU Joseph Myers ended up doing 15-20 test conversions for people to comment on and compare to Maxim's conversions (which were also updated several times in response to issues found by the reposurgeon people). Apparently, there were fixed costs of roughly 4-5 hours spent in git alone IIRC, so each test repo conversion took at least 5 hours on whatever machine it was that Joseph used.

Here's hoping things go smoothly over the weekend...

[pal666]

simultaneously pretending lost commit history is just an unimportant detail.

history isn't lost, you have your readonly svn clone for all eternity

[Teggs]

...real server class hardware and not in someone's basement.

To be fair, Michael has a 7742 EPYC server in his basement.

[smitty3268]

If they actually meant to get it done, then why was it delegated to ESR in the first place? Assigning projects to people who are lazy and/or incompetent is a well known strategy used by managers who wan't to kill a project, but don't want to do it outright.

Nobody cared 2 years ago. About 2 months ago they decided they needed to get it done and a lot of manpower got assigned.

[caligula]

This has to be fake news! No way ESR could get something that big done this quick! He's not procrastinated nearly enough so this has got to be an out-of-season April fool's joke or ESR being replaced by a shapeshifting alien or a pod person.

Maybe he got a free new quad-socket 128 core EPYC machine from AMD.