- Apr 01, 2021
-
-
Raphaël Gomès authored
This potentially simplifies the logic and removes a bug (thankfully only theoretical) from `gitlab_special_ref_target`
-
- Nov 01, 2024
-
-
Georges Racinet authored
-
Georges Racinet authored
-
- Oct 31, 2024
-
-
Georges Racinet authored
-
Georges Racinet authored
-
Georges Racinet authored
(forgot to commit for 2.5.4)
-
Georges Racinet authored
-
Georges Racinet authored
-
- Oct 30, 2024
-
-
Georges Racinet authored
Closes #162
-
Georges Racinet authored
Up to know, we had a divergence between conversion of responses and requests: the former were able to convert deep in the message, with path expressions and iteration over lists. The later was able to convert more general revspecs than just commit SHAs. We now use the deep conversion facility, `attr_path_to_git`, for both, and hence give it the general revspec converter. Because responses are gRPC messages while requests are just dicts (at this stage of treatment), we had to pass object handling lambdas to `attr_path_to_git`.
-
- Oct 31, 2024
-
-
Georges Racinet authored
This will reduce the boilerplate and treats a few traps (mutation for non-scalar fields)
-
Georges Racinet authored
It was a bit weird to have something called a normalizer and apply it only to the Gitaly error. And it will be simpler indeed for next improvement to have it apply to both received errors.
-
- Oct 28, 2024
-
-
Georges Racinet authored
Users expect the message (that they had to write in the MR squash dialog) to be applied whether this is an actual fold or not. Something similar was done in the Git case by upstream GitLab. See heptapod#1846
-
- Oct 22, 2024
-
-
Georges Racinet authored
Upstream introduced the `find_renames` flag, that triggers returning a change as a rename if needed, together with the old path. We do not perform a comparison test in the `COPIED` case, as we couldn't get Gitaly to issue some (nor `git diff-tree` for that matter). Gitaly tests themselves avoid the subject. Closes #206
-
- Oct 26, 2024
-
-
Georges Racinet authored
Unfortunately, the case of renames (result of `hg mv`) was not tested at all. We do not add a test for this, because next changeset with (together with its related features).
-
- Oct 23, 2024
-
-
Georges Racinet authored
-
Georges Racinet authored
-
- Oct 22, 2024
-
-
Georges Racinet authored
Ambiguity about whether it is before of after application of pagination parameters did not resist Comparison tests for long Closes #216
-
- Oct 17, 2024
-
-
Georges Racinet authored
-
Georges Racinet authored
-
Georges Racinet authored
-
Georges Racinet authored
The comparison tests raised two problems: 1. `UserFFBranch` now returns structured errors in case of mismatch in `expected_old_oid`. Funnily enough, it does not fill the `reference_name` error field 2. `SearchFilesByContent` does not normalize line endings the Unix way any more, but it still adds `\n' at the end if missing. To solve this, we include the line ending in the regexp group and treat the missing ending separately
-
- Oct 15, 2024
-
-
Georges Racinet authored
Since Heptapod does not support Mercurial signature at all, the main motivation is that the Unimplemented error could become a hard error at some point. The comparison test may seem trivial, but it actually proves that no error and the empty string are the expected response for Gitaly servers without a configured signing key, hence HGitaly may and will simply look one of those.
-
- Oct 09, 2024
-
-
Georges Racinet authored
-
Georges Racinet authored
-
Georges Racinet authored
-
Georges Racinet authored
In case a different name branch is to be merged and it makes up a linear history, we need to perform a merge anyway, similar to what happens with the `semi-linear` method. Actually it needs to happen also with the semi-linear method, despite the request not being considered a fast forward any more. See heptapod#1823
-
Georges Racinet authored
At this point, it is best to have a couple booleans and express their logical connections.
-
- Sep 30, 2024
-
-
Georges Racinet authored
The default Heptapod series will now proceed to 17.5, hence we will bump the minor version.
-
Georges Racinet authored
-
Georges Racinet authored
-
Georges Racinet authored
-
Georges Racinet authored
No actual content.
-
- Sep 25, 2024
-
-
Georges Racinet authored
-
Georges Racinet authored
-
Georges Racinet authored
This part was forgotten in 2.3.2, hence it internally claims to be 2.3.2dev0 which resulted in wrong RHGitaly uploads.
-