Download less of all-cabal-hashes

[Blaisorblade]

We can't just download the head of all-cabal-hashes, since a snapshot can request old revisions of Cabal files by hash (see #2175 and #2464).

But as mentioned, downloading the entire repo's history is a long-term problem.
Possible solution I proposed:

I imagine one could add tags for each snapshot, and stack could fetch those tags when needed, but the tags would be needed remotely.

[mgsloan]

Cool idea!

[Blaisorblade]

@mgsloan Thanks! But to be clear: I plan on leaving this to somebody familiar with all the projects and having the necessary rights.

[Blaisorblade]

I probably should have pinged @snoyberg on this ages ago.

[Blaisorblade]

Now we should have the info we need—it's not tags but a snapshot field:
fpco/stackage-curator@67484d7

[Blaisorblade]

@borsboom I figured this is probably too hard to fix safely for 1.2.0. The best we could do in that timeframe is revert the fix for #2175 for now—that would keep the existing unreproducibility and cause no behavior regression compared to 1.1.2.
Also, the fix would only affect new snapshots. I also confess I'm short on time, which doesn't help.

Needed steps for a fix:

• parse extra info from snapshot (easy);
• get the information from the parsed snapshot to the repo checkout (nontrivial for me, because I'm not familiar with the code yet);
• in fact, this might require fetching new info for each snapshot, which requires restructuring the way this info is fetched; that might be genuinely nontrivial.
• actually use extra info for checkout (finding the right git command is easy).

I hear @mgsloan might have more insight though?