- Oct 21, 2022
-
-
Georges Racinet authored
We already know that there will be changes in protocol in the default branch.
-
Georges Racinet authored
-
Georges Racinet authored
Release needed for 0-34-testing Docker image to work. Also regenerated the Ruby lib, which changed as expected only the version.
-
- Oct 20, 2022
-
-
Georges Racinet authored
Depending on coverage and maybe Python version, they are correctly accounted for or not
-
Georges Racinet authored
What we need right now is obsolete information, we're doing a bit more by including parents. Beware that obsolescence needs both `UNFILTERED` to be specified (otherwise there's currently an uncatched exception) and the `OBSOLETE` field to be required (otherwise it will be unset, which looks to be `False` with gRPC if one does not explicitely check that it is set). Error treatment is really minimal. Cases of lookup error happen when the revset itself cannot be resolved (not if no changesets match), hence we return INVALID_ARGUMENT in that case, using the style recommended by #99. Closes #110
-
Georges Racinet authored
-
Georges Racinet authored
This new service will carry Mercurial-specific methods involving changesets, listing them with specific information that has no Git counterpart and more precisely #110 being the first use case. As it felt really weird to define a Mercurial changesets as having just an `id` and an `obsolete` flag, we defined more fields. Those that already exist in the Git case will probably be the last ones to be implemented. We need the `MercurialRepositoryView` enum for the same reason that `hg log` needs the `--hidden` flag even to display information about a single obsolete changeset: `MercurialListChangesets` takes any revset and switching straight away to the unfiltered repo would change the meanings of symbolic handles (branch or topic name, etc.)
-
Georges Racinet authored
Doesn't harm if we generate the library with the same Ruby version.
-
Georges Racinet authored
-
- Oct 19, 2022
-
-
Georges Racinet authored
Nowadays, in mercurial-testhelpers, it's `commit_file()`.
-
Georges Racinet authored
-
Georges Racinet authored
Migrated two tests to that style, one with empty repo the other with the tree shaped factory. Hoping it'll make tests a tad bit more focused and readable
-
Georges Racinet authored
In the process of migrating some tests in `test_commit.py`, realized that some of them were using `make_tree_shaped_repo`, hence passing it as a factory function would be desireable. It would have needed having `make_empty_repo` return a third value (an empty dict of changesets), so instead of modifying hundreds of tests, we made a new variant.
-
Georges Racinet authored
This prevents the silly dance of overridding `__init__()`, even though we have to use `attr.ib()` in the case of `RepositoryService`
-
- Sep 21, 2022
-
-
Georges Racinet authored
-
Georges Racinet authored
-
Georges Racinet authored
-
Georges Racinet authored
-
- Jul 18, 2022
-
-
Georges Racinet authored
-
Georges Racinet authored
-
Georges Racinet authored
Surprisingly, Gitaly just returns an empty stream in that case. Closes #104
-
Georges Racinet authored
Things were not as clear as they are now when this was written. Now the contract of `gitlab_revision_changeset` is to input the filtered repo, and it will look into the unfiltered repo when it needs to.
-
Georges Racinet authored
As the comment suggested, it was not necessary in this case, and I am still wary of it (also lots of useless code running). The resulting code would have been directly compatible with Mercurial before and after 5.5
-
Georges Racinet authored
We are now requiring at least 6.1 (through py-heptapod)
-
Georges Racinet authored
If the passed revision is not found, that is probably an error from the caller, it's worth more that `DEBUG`
-
Georges Racinet authored
In particular, py-heptapod 4.0 (heptapod as a setuptools project name) requires Mercurial >= 6.1
-
Georges Racinet authored
We will use version 15.0 of the protocol, update minimal requirements etc.
-
Georges Racinet authored
As usual a direct hash or a keep-around could be obsolete
-
Georges Racinet authored
Closes #100 As displayed in the test and in the issue logs, even a direct sha wasn't supported. The complication here is that we have the parsing logic for the `..` and `...` Git revspec operators. Hence the parser function must return the new repo view to use.
-
Georges Racinet authored
I usually prefer early returns to repetition of boolean conditions, as it makes really sure that the work is done and helps keeping later code simpler. Some people disagree.
-
Georges Racinet authored
In particular, in CI, it will display immediately the offending lines (typically just a few of them).
-
Georges Racinet authored
The point with special refs is that they often point to obsolete changesets (keep-arounds, merge requests)
-
- Jul 08, 2022
-
-
Georges Racinet authored
-
- Jun 23, 2022
-
-
Georges Racinet authored
-
Georges Racinet authored
-
Georges Racinet authored
One minute was too low for foss.h.n, creating disruption in backups. Ideally this should be made configurable (no time to do it right now).
-
- Jun 22, 2022
-
-
Georges Racinet authored
-