1. 12 May, 2018 2 commits
  2. 11 May, 2018 15 commits
  3. 05 May, 2018 1 commit
  4. 04 May, 2018 5 commits
  5. 02 May, 2018 1 commit
  6. 30 Apr, 2018 1 commit
  7. 28 Apr, 2018 3 commits
    • appveyor: Re-enable non-tag builds · 741b7052
      Kai Willadsen authored
      Appveyor has some issues building tags if the tag isn't actually the
      most recent commit on that branch. In our release workflow this is
      theoretically fine, but because we're mirroring the repos across to
      Github, in reality I've always pushed the post-release version bump
      commit before the tag actually gets mirrored.
    • Revert maint rubbish commit · e3572d7a
      Kai Willadsen authored
      Well I'm glad my idiocy is immortalised forever like this. This is good.
      I'm happy right now.
    • setup_win32: Get site-packages in a slightly less silly way · e5c572da
      Kai Willadsen authored
      This was breaking on msys2 builds, because it only has one entry in
  8. 27 Apr, 2018 2 commits
  9. 25 Apr, 2018 1 commit
    • sourceview: Fix pasting multi-byte encodings (#179) · 4b6f3652
      Kai Willadsen authored
      For this API, pygobject isn't handling string length arguments for us,
      so we had to do it ourselves. Shockingly, I got it very wrong. The API
      requires a UTF-8 encoded string (which pygobject transparently encoded
      for us) but the length arg has to be the bytestring length, not the
      unicode codepoint length, which is what we were previously passing
  10. 22 Apr, 2018 1 commit
  11. 21 Apr, 2018 1 commit
  12. 20 Apr, 2018 7 commits
    • misc: Fix translation markings · 663a0552
      Kai Willadsen authored
    • filediff: Add a helper for showing user-criticial errors · db57bf5b
      Kai Willadsen authored
      The idea behind this decorator is that it should be used with functions
      where failure is important to the user (and may not be obvious
      otherwise). The only current use is on save, where it's possible to hit
      a traceback and have the callback handling crash out, but other than a
      log message or traceback on the command line... the user may not notice
      This is made necessary by the GTK+/pygobject philosophy (which is
      generally extremely good) of treating errors of this kind as non-fatal.
      We also want it to not be fatal in this case, but... the user needs to
      know if they tried to save a file and it didn't work.
      On the other hand, if they tried to e.g., go to the next diff and we
      messed something up and didn't go to the next diff, (a) they'd notice
      and (b) showing them a modal dialog would be even more annoying than
      the actual error.
    • FIXME: Initial pass at a wrapper that ensures that a user sees that we've… · 88d511ab
      Kai Willadsen authored
      FIXME: Initial pass at a wrapper that ensures that a user sees that we've errored, for certain critical things (like saving...)
    • filediff: Work around GtkSourceView load failure (final part of #152) · da3ba746
      Kai Willadsen authored
      When GtkSourceView's loader fails to load due to a unicode decode issue,
      at least one of the failure cases doesn't actually end the in-progress
      user action (see bgo#795387). Because we use user-actions to handle both
      our undo stack and also our sensitivity setting (i.e., we disable some
      things while you're in the middle of a user action) this breaks several
      things across the Meld UI, basically putting us in a very weird state.
      The workaround here is... not good, but it's the best we can do for now.
      If/when the GtkSourceView side of this gets fixed, we'll end up with a
      lot of GTK+ critical logs, but everything *should* continue to work.