1. 07 Oct, 2020 1 commit
    • Sietse Brouwer's avatar
      ngtip: fix TypeError: make revset from revs, not nodes (issue6405) · e25394b6d021
      Sietse Brouwer authored
      This commit does two things:
      
      - Firstly, it tweaks the ngtip revset test to exercise the -G/--graph
        flag. This successfully triggers the bug when it is present.
      - Secondly, it changes the `ngtip` revset to return a revset made from
        integer revs instead of node hash bytes. The test now passes.
      
      Details: The TypeError was triggered by running
      
          hg log -r 'ngtip("default")' --graph
      
      in a repository with more than one changeset. The --graph tag caused the
      flow of control to call `reachableroots2` with the changeset ID found by
      the `ngtip` revset. Because the changeset ID was a node hash (bytes)
      instead of a rev (int), reachableroots2 raised the following error:
      
          TypeError: an integer is required (got type bytes)
      
      --HG--
      branch : stable
      e25394b6d021
  2. 12 Oct, 2020 1 commit
  3. 18 Sep, 2020 1 commit
  4. 26 Aug, 2020 4 commits
  5. 17 Sep, 2020 3 commits
  6. 10 Sep, 2020 1 commit
  7. 05 Sep, 2020 2 commits
    • Matt Harbison's avatar
      config: specify the extension when forcing a config option · 1ee535d769db
      Matt Harbison authored
      I noticed that `hg config server --debug` said 'none: server.bundle1=False'.
      
      --HG--
      branch : stable
      1ee535d769db
    • Matt Harbison's avatar
      tests: use the modern config setting to enable evolution in topic tests · 044ce00dfa88
      Matt Harbison authored
      When adding `hg pick --config extensions.evolve=` to test-topic-dest.t to
      investigate issue6406, it failed with `abort: unknown command 'pick'`.  This is
      a manifestation of issue6039.  I made several attempts at determining the root
      cause, and while I'm still worried about *why* a config setting can cause the
      extension to fail to fully initialize, there are some topic bugs to cleanup and
      I don't feel like spending more time on this.
      
      --HG--
      branch : stable
      044ce00dfa88
  8. 08 Sep, 2020 3 commits
  9. 05 Sep, 2020 1 commit
  10. 26 Aug, 2020 1 commit
    • Martin von Zweigbergk's avatar
      dirstate: replace _uncommitdirstate() by core's movedirstate() · 2df762b9c0c5
      Martin von Zweigbergk authored
      IIRC, when _uncommitdirstate() was upstreamed, it got cleaned up a
      lot. Let's switch to the simpler upstream version.
      
      This fixes the brokeness in `hg uncommit` demonstrated by the previous
      patch. The problem was that _uncommitdirstate() had not been properly
      updated to support uncommitting to another commit. If you replace
      `oldctx.p1()` by `ctx` there, you would get this same diff as from
      this patch (the `interactive` code path had been fixed this way
      already).
      
      --HG--
      branch : stable
      2df762b9c0c5
  11. 03 Sep, 2020 1 commit
    • Martin von Zweigbergk's avatar
      tests: add test showing that `hg uncommit --rev` can fail to mark file added · 80b5d4b85a52
      Martin von Zweigbergk authored
      The test case modified in this commit uses `hg uncommit --rev 0 b` to
      revert the parent commit's state to revision 0 where file `b` doesn't
      exist. The file is still tracked in the dirstate (correctly, since `hg
      uncommit` should not affected the tracked-ness), so `hg status` should
      report it as added. However, as the calls added in this patch shows,
      it gets reported as modified instead.
      
      --HG--
      branch : stable
      80b5d4b85a52
  12. 02 Sep, 2020 1 commit
    • Matt Harbison's avatar
      topic: fix the `topic(...)` revset to not select commits without a topic · 95478db35f88
      Matt Harbison authored
      The problem seems to be that the now removed __contains__ check would select any
      revision indicated by a non-string argument.  This led to `topic(head())`
      returning a bunch of non-topic heads, though not as many as simply `head()`.
      
      I'm likely misunderstanding something, because there appears to be a similar
      __contains__ check for `branch()` in core.  But maybe that's just harder to see
      the issue, because branch names doesn't disappear like topics.
      
      --HG--
      branch : stable
      95478db35f88
  13. 21 Aug, 2020 4 commits
  14. 02 Aug, 2020 1 commit
  15. 31 Jul, 2020 3 commits
  16. 12 Jun, 2020 1 commit
  17. 11 Jun, 2020 1 commit
    • Anton Shestakov's avatar
      topic: treat argument to --topic as a literal topic name explicitly · 7f80558c9b8d
      Anton Shestakov authored
      We already do treat it as a string (i.e. not a special revset language symbol)
      that needs to be escaped, because we use %s in repo.revs(), which escapes and
      single-quotes the appropriate argument. But now we abort in case user supplies
      topic name that doesn't exist.
      
      Before this patch, when user ran something like `hg push --topic .`, this code
      would use 'topic(".") - obsolete()' revset, which would get an empty set from
      repo.revs() unless there was a topic named "." (unlikely, because this name is
      reserved). This code would then add an empty set to the 'rev' argument of the
      original hg push command. And `hg push` without -r will try and push
      everything, and not just changesets on a specific topic.
      
      --HG--
      branch : stable
      7f80558c9b8d
  18. 27 Jul, 2020 1 commit
  19. 22 Jul, 2020 7 commits
  20. 21 Jul, 2020 1 commit
  21. 04 Jul, 2020 1 commit