I checked out this branch, installed cabal-install from it and reran a timing test against amazonka, same as #10605 (comment). This gives slow results. I ran it twice after cleaning and got 82s and 84s. I reran the same test twice using cabal-install from the master branch, at 1f52963, and got times of, 16s each time.

@philderbeast Yes, that's what I'm expecting. It might be possible to keep the shallow clone behavior in some cases without introducing regressions, but let's fix the regressions first.

I've commented this in #10605 (comment), but to re-iterate here:

It seems to me that we can easily fix the regression by only reverting to the previous behaviour (full clones) when the hash is not given in full. For all other users using the full hash, we keep the happy fast path.

@alt-romes That seems reasonable, although I do worry (as with all roll-forward fixes) that we'll be introducing another bug or missing some edge case as we do so. Would you put up a PR for it?

@alt-romes That seems reasonable, although I do worry (as with all roll-forward fixes) that we'll be introducing another bug or missing some edge case as we do so. Would you put up a PR for it?

I can appreciate that concern, and still think the cost of being that conservative here seems quite high. Since the regression seems well define, let's try fixing it specifically. I'll make a PR, including your regression test.

Closing in favor of #10639