summaryrefslogtreecommitdiff
Commit message (Collapse)AuthorAge
* dgit(7): Mention bad origs as possible cause of quilt fixup failureIan Jackson2018-08-20
| | | | | | And document our error message changes. Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* dgit: quilt fixup error messages: Change hint emphasisIan Jackson2018-08-20
| | | | | | | | | * No longer hint --quilt=smash or dpkg source --commit. These are rarely the right thing. * Hint about bad origs, instead. This is fairly common. Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* dgit: Remove a spurious spaceIan Jackson2018-08-20
| | | | Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* test suite: gdr-unprocessable-hints: New testIan Jackson2018-08-20
| | | | Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* git-debrebase: Provide hints for unprocessable commitsIan Jackson2018-08-20
| | | | | | | | | | Depending on the apparent branch ffq state. Closes:#905005. In particular, if it doesn't look like this branch was ever used with gdr, suggest that the user is on the wrong branch, or needs to use convert-from-*. Closes:#905279. Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* git-debrebase: Introduce new `fail_unprocessable'Ian Jackson2018-08-20
| | | | | | | | | This is called when we want to call `fail' because we have found an unprocessable commit. NFC. Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* git-debrebase: convert-from-*: leave debrebase-last refsIan Jackson2018-08-19
| | | | | | | To hint to everyone that this is now a gdr branch. This will make dgit push use gdr make-patches, for example. Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* git-debrebase: begin_convert_from: Break out $gdrlast_objIan Jackson2018-08-19
| | | | | | NFC Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* git-debrebase: Snag on repeated conversionsIan Jackson2018-08-19
| | | | | | | | | If debrebase-last says we are ahead of some previous gdr work, it is carrying on is unlikely to be helpful. It would only be sensible if the package is vacillating between branch formats, and this is reflected and visible in the branch history. Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* git-debrebase: Refuse to start conversion if ffq-prev existsIan Jackson2018-08-19
| | | | | | | Trying to rerun git-debrebase convert-from-something, on an unstitched gdr branch, would end in tears. Forbid it utterly. Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* git-debrebase: record_gdrlast: Honour $oldvalueIan Jackson2018-08-19
| | | | | | NFC for existing callers. Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* git-debrebase: Move ffq_prev_branchinfo up the fileIan Jackson2018-08-19
| | | | | | NFC Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* git-debrebase: Break out begin_convert_from & complete_...Ian Jackson2018-08-19
| | | | | | | | $gdrlastinfo is undef for now, and will be used in a moment. No functional change. Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* git-debrebase: Break out record_gdrlastIan Jackson2018-08-19
| | | | | | NFC Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* git-debrebase: get_tree: Check we got an object nameIan Jackson2018-08-19
| | | | | | Otherwise the cache is dangerous. Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* git-debrebase: Merge: Make a todo about gbp not round trippingIan Jackson2018-08-19
| | | | Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* git-debrebase: Merges: Enable with command-line option instead of env varIan Jackson2018-08-19
| | | | Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* git-debrebase: New walk mechanism: Drop obsolete $opt_carefulIan Jackson2018-08-19
| | | | | | NFC Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* git-debrebase: Merge: Put off some xxx's, making them MERGE-TODOIan Jackson2018-08-19
| | | | | | | We are going to release with this machinery in this experimental state. Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* git-debrebase: Document merged-breakwater metadata tagIan Jackson2018-08-19
| | | | | | | This is generated only by the experimental merge machinery and is subject to possible future change. Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* git-debrebase: Merge: Remove some xxx'sIan Jackson2018-08-19
| | | | | | | | | | git-debrebase(5) already has ILLEGAL OPERATIONS (which forbids many git-rebase operations) and also of course LEGAL OPERATIONS. The limitations on what kind of merges we can resolve and how will be documented when we document the merge machinery properly. Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* changelog: Document changes to convert-from-gbp messages.Ian Jackson2018-08-19
| | | | | | This is part of #906641. Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* git-debrebase: convert-from-gbp: Check separately for $dtagIan Jackson2018-08-19
| | | | | | | This allows us to print a more sensible error message, Previously git would print a fatal error, after which we carry on. Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* git-debrebase: convert-from-gbp: Add a missing newline to a dieIan Jackson2018-08-19
| | | | Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* git-debrebase: convert-from-gbp: handle missing maintainer tagIan Jackson2018-08-19
| | | | | | | | Look for / in $mtag, not in $mtag_pat. Obviously it will be in the latter. The result, before, is that we would look for a completely bogus $dtag and print daft error messages. Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* test suite: gdr-import-nostitch: New testIan Jackson2018-08-19
| | | | | | | Generate this situation and check it's OK. Also check that the messages contain the expected hint. Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* git-debrebase: convert-from-gbp; on stitch failure, clarify messageIan Jackson2018-08-19
| | | | | | | | | Clarify what the preceding failure messages mean, and what the consequences are. This is part of #906641. Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* test suite: gdr-convert-gbp: Break the start out into lib-gdrIan Jackson2018-08-19
| | | | | | | This will allow us to test other things to do with this conversion without making tests/setup/gdr-convert-gbp more complicated. Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* git-debrebase: convert-from-gbp: Print a success message.Ian Jackson2018-08-19
| | | | | | This is part of #906641. Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* changelog: Add close note for #905400 to changelog entry for 6.5.Ian Jackson2018-08-19
| | | | | | | | The commit git-debrebase: Properly reject bare dgit dsc imports. fixed #905400. Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* test suite: gdr-convert-gbp: test snag on unused patchesIan Jackson2018-08-19
| | | | Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* git-debrebase: convert-from-*: Snag on unused patchesIan Jackson2018-08-19
| | | | | | | | | | | Patches in d/patches which are not in series will be deleted, and not represented in the output. So call them a snag. Closes:#904997. Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* git-debrebase: get_tree: Support $precheck and $recurseIan Jackson2018-08-19
| | | | Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* test suite: gdr-subcommands: Test raw conversion from gbpIan Jackson2018-08-19
| | | | | | | | | Make a branch which strips off the pseudomerge generated by gdr to stitch the dgit history, and then run some tests on that. These new tests detect #905975. Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* test suite: gdr-subcommands: Use $peel, not $work, for a few checksIan Jackson2018-08-19
| | | | | | | | | | | | The definition of P says P result is add-patches, remaining letters apply to result~ But this was not hnoured by N n Z. Honour it. No change with existing tests, which do not follow P with any of those. But, this will be important in a moment, Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* test suite: gdr-convert-gbp: Save converted-from-gbpIan Jackson2018-08-19
| | | | | | This new branch will be useful in a moment. Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* git-debrebase: On rebase, always save ffq-prevIan Jackson2018-08-19
| | | | | | | | Even if we were both stitched and laundered. Closes:#905975. Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* git-debrebase: `prepush' is now a silent no-op if unstitched.Ian Jackson2018-08-19
| | | | | | This is more friendly. Update the docs accordingly. Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* changelog: Add perf numbersIan Jackson2018-08-19
| | | | | | | | | | | | | This is from tests on my laptop with the branch the Debian kernel folks shared with me, ie c6eb63c6ed0694b0d22fbe5aaff953d209a3fc4e. avoid new runtimes in [s] before read-tree get_differs git-debrebase 78.574 3.699 1.419 git-debrebase status 3.359 unchanged 0.818 Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* git-debrebase: Rewrite get_differs in terms of trees_diff_walkIan Jackson2018-08-19
| | | | | | | | | | | As noted, git-diff-tree is quite slow. We can do much better, by stepping across the two trees ourselves. There should be no particularly significant change: the answers should be the same for reasonable inputs. Unreasonable inputs (particularly, executable files etc.) may be handled better, even. Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* git-debrebase: get_tree: Support '' for nonexistent objectsIan Jackson2018-08-19
| | | | | | This is going to be convenient in a moment. Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* git-debrebase: New get_tree machinery: Fix debuggingIan Jackson2018-08-19
| | | | | | | | | When we localise, we get a new variable, rather than localising the binding imported from Debian::Dgit. How annoying. Also, add two new debug prints at level 3. Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* git-debrebase: Fix comment documenting D_DEB_*Ian Jackson2018-08-19
| | | | Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* git-debrebase: Actually avoid read-tree in walk when not rewritingIan Jackson2018-08-19
| | | | | | | | | | | | | | | | | | | | See git-debrebase: Prep for avoid read-tree in walk when not rewriting for the motivation. Here we try to check that the commit we would make, if we were actually calling read-tree, write-tree and hash-object, would be identical to the one we are reusing. In particular, we check that the upstream files, and the debian/ directory, have the contents we would feed to read-tree. And we check that the parent list is the same. With these checks this new algorithm is, I think, safe enough to just use always. So set $opt_careful to 0. We will abolish this variable in a moment. Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* git-debrebase: Provide new get_tree and trees_diff_walkIan Jackson2018-08-19
| | | | | | | | | | | | | | These are ways to avoid calling git-diff-tree, which is very slow even with --name-only --no-renames and without recursion. I think it is doing all the work to make nice actual diffs, and then throwing them away. Two calls to git-ls-tree is about 20x faster than git-diff-tree on the Linux kernel, for example. No functional change in this commit, since no callers yet. Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* test suite: unset GIT_EDITOR, so it works if user has that set.Ian Jackson2018-08-19
| | | | Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* git-debrebase: Rename $want_tree_ to $want_Ian Jackson2018-08-19
| | | | | | NFC Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* git-debrebase: Prep for avoid read-tree in walk when not rewritingIan Jackson2018-08-19
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Because the git index is a flat list of files, not a directory hierarchy, git-read-tree can be very slow. We want to avoid it if possible. Introduce a new way of working, where we defer calls to read-tree. In detail: * $read_tree_upstream and $read_tree_debian, which are the only ways that walk invokes read-tree, simply record their argument; * The actual read-tree is done just before the write-tree and commit generation. read_tree_upstream, conveniently, does both halves, if it gets both arguments. * Put all of the read-tree and commit regeneration in a branch triggered if we are rewriting or "trying to be careful". Right now we are always careful, but I have tested this with a hack setting $opt_careful to 0, and the performance on the example branch in #905995 is much improved: 3.4s rather than 77s for `git-debrebase'. This is not suitable for enabling in its current form. I want to actuallly abolish the $opt_careful, and, instead, check with get_differs or git-cat-file that the tree segments, and the parent lists, are identical. When we do that we won't be checking any more that the commit generation can generate identical commits when not rewriting. But it doesn't really matter much any more provided the commits are well-formed and right, and the test suite will check that. Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* dgit(1): Encourage --overwrite rather than --overwrite=version.Ian Jackson2018-08-19
| | | | Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
* git-debrebase: walk: Introduce $read_tree_debianIan Jackson2018-08-12
| | | | | | We are going to make these do something more efficient, sometimes. Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
generated by cgit v1.2.3 (git 2.25.1) at 2024-04-30 17:37:46 +0000