3 subject="""comment 2"""
4 date="2020-04-28T20:56:01Z"
6 Fairly sure it's something to do with the chunk log. The tests with no
7 chunks succeed, so the subsequent failures could be due to it failing to
8 see information about chunks used to store the key when it then tries to
11 Aha: It's using Annex.eval, so it does stuff, and throws away the changed
12 state. So, after writing to the journal, when journalIgnorable = False,
13 it will then forget that, and will try to read from the branch, ignoring
16 I checked all Annex.eval, and all the rest have the pattern
17 Annex.eval =<< Annex.new. This is the only one that *reuses*
18 an old state across multiple Annex.eval. And it's not like the commit that
19 triggered this made Annex.eval more dangerous; that were never a good thing
20 to do and probably only didn't break on some other state being lost before