path: root/dgit
Commit message (Collapse)AuthorAge
* Split brain: TODOsIan Jackson2016-09-05
| | | | Signed-off-by: Ian Jackson <>
* Split tags: Push the maintainer view tag, where supportedIan Jackson2016-09-05
| | | | | | | | | | | | | | | | | We introduce a new `maint' possibility in to the dgit-tag-format access cfg list, presence of which indicates that the receiver is content to receive DEP-14-style maintainer-view (ie, non-dgit0 tags. For now, we default this to true. However, repos handled by dgit-repos-server cannot currently cope with this, because: * dgit-repos-server currently tries to be compatible with old dgits which send DEP-14-named dgit tags; * dgit-repos-server doesn't currently permit the creation of `extraneous' (ie, non-dgit) tags and branches. So for Debian, we do not set this flag. Signed-off-by: Ian Jackson <>
* Split tags: Genrate maintainer-view tag tooIan Jackson2016-09-05
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | push_tagwants gets a new argument, $maintviewhead, which is defined iff the quilt mode means we want split tags. It then specifies the generation of both tags. push_mktags gains the ability to make the new `maint' view tag (and checks that the view is not something else unexpected). Introduce a function debiantag_maintview for calculating the maintainer's idea of the tag. Currently we use DEP-14 encoding (and this function is a clone-and-hack of, but perhaps we should do something more complicated (depending on the quilt mode). For now we do not push the maint view tag anywhere. The ability to do that, where appropriate, will come in a moment. This commit introduces protocol version 4, which includes the new `maint-view' param. When have a split brain quilt mode, we need to be sure that we're using protocol version 4 or out peer may not honour this, resulting in an annoying failure later (the responder sending a different number of tags to the number expected). We also have to check that we're using the new tag format (or we might want to try to generate two different tags with the same name, which is madness). When the quilt mode is not a split brain one, this whole commit should produce no overall functional change (even though a higher protocol version may be negotiated). Signed-off-by: Ian Jackson <>
* Split tags: Preparation: Rename tag variables etc. in dopushIan Jackson2016-09-05
| | | | | | | | | | | | | | We are going to have different branches at once. Rename some variables to permit this distinction. We have $actualhead (whatever HEAD is), $dgithead (the dgit view), and $maintviewhead (the maintainer's view if different, or undef if not0. Change the existing references. Also change one reference to HEAD to $dgithead: when we update the local tracking ref for the dgit view, we need to use $dgitview. Signed-off-by: Ian Jackson <>
* Split tags: Preparation: Reorganise tagwants and mktagsIan Jackson2016-09-05
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | We are going to want to generate two tags. The current code structure is not really set up for this. Also the knowledge of what tags are being made needs to be used both in dopush and in the push responder. So: * Introduce push_tagwants, which calculates which tags we are going to be making, including some details of them. * `View' indicates which tag this is. For now there is only one, `dgit'. But the tags are going to be in a defined order - or, at least, the `dgit' tag will come first. * Have push_parse_changelog no longer return the single tag name, and abolish the corresponding $i_tag varaiable. * push_mktag bcomes push_mktags and takes much of its instruction from @$tagwants. * push_mktags does more than just making tags - it also updates the dsc, and does some checks, so it needs to fish the relevant object id out of the dgit view tag. * $mktag in push_tagwants can now operate on a tagwant, and the tag-making part of push_mktags just iterates over the tagwants. * The filename plumbing in dopush is somewhat generalised: we collect the filenames out of push_mtags or responder_receive_files, and then substitute them into the tagwants. * The tag checking iterates over the tagwants. * The push spec computation iterates over the tagwants. Signed-off-by: Ian Jackson <>
* Split tags: Preparation: Break out (and fix) need_tagformatIan Jackson2016-09-05
| | | | | | | | | | No major functional change. However, we do preserve the previous override value, so that need_tagformat's caller is happy, we still override the access configuration. Signed-off-by: Ian Jackson <>
* Split tags: Preparation: Break out access_cfg_tagformatsIan Jackson2016-09-05
| | | | | | No functional change. Signed-off-by: Ian Jackson <>
* Split tags: Preparation: move select_tagformat into a new sectionIan Jackson2016-09-05
| | | | | | No functional change. Signed-off-by: Ian Jackson <>
* Split tags: Introduce $mktag in push_mktagIan Jackson2016-09-05
| | | | | | | | | | | | | | push_mktag is going to want to become push_mktags, and make perhaps two tags. For now, break the actual tag-making out into a subref. Rename the arguments to push_mktag to ensure we caught all the references in $mktag. push_mktag now wants to return an array; make sure that the call sites all give it array context. Signed-off-by: Ian Jackson <>
* Tag change: Double check agreement in rpush protocolIan Jackson2016-08-14
| | | | | | | Send the tag format as a param and check that it is the same at both ends. This check _ought_ not ever to fail, but I want to be sure. Signed-off-by: Ian Jackson <>
* Tag change: Rename $tagformat to $tagformat_wantIan Jackson2016-08-14
| | | | | | No functional change. Signed-off-by: Ian Jackson <>
* dgit: Use GIT_COMMITTER_DATE for automatic patchIan Jackson2016-08-14
| | | | | When automatically generating quilt patch, honour GIT_COMMITTER_DATE for filename creation (makes filename deterministic in test suite).
* Tag change: Calculate desired tag formatIan Jackson2016-07-31
| | | | | | | | | | | | Introduce machinery for deciding what format tag to make, during push. The computation is slightly ad-hoc. Currently it always returns `old' and does as before, so no overall functional change with existing configs and existing command lines. Reserve protocol version 4 for new tag support. Signed-off-by: Ian Jackson <>
* Tag change: dgit: Fetch both old ane new format tagsIan Jackson2016-07-31
| | | | | | We are going to want to know about all of these. Signed-off-by: Ian Jackson <>
* Tag change: Rename debiantag to debiantag_oldIan Jackson2016-07-31
| | | | | | | | | | | | | | | | | We are going to change the tag name of the main signed tag used by dgit to signal the user's intent to the archive server. Prepare for this by changing debiantag to debiantag_old and changing all call sites outside dgit itself. dgit will want to be able to generate either, depending on whether it's cooperating in an rpush with a new or old dgit so reintroduce a wrapper called debiantag there. This will also allow us to have dgit generate old tags for testing. No functional change. Signed-off-by: Ian Jackson <>
* dgit: rpush protocol comment: Add some missing protocol elementsIan Jackson2016-07-31
| | | | | | | progress, supplementary-message, previously, and param csuite, were all missing. Document them. Signed-off-by: Ian Jackson <>
* Split brain: When pushing, find the dgit view in the cacheIan Jackson2016-07-17
| | | | | We now do set $split_brain, and we don't do anythig with the cached dgit view, so we'll crash at the die, in a moment.
* Innards: Improve mktree_in_ud_from_only_subdir messageIan Jackson2016-07-17
| | | | When it finds multiple subdirs, print them.
* Subprocess error handling: Initialise $? to -1Ian Jackson2016-07-17
| | | | | | | | | | | | | | | When system(3perl) fails due to syscall error, it sets only $!. When it succeeds it sets only $? and sometimes trashes $!. Conversely, close of a popened filehandle always sets both in all cases. Document this in a comment. So when using system and relying on $?/$! (rather than looking at system's return value), such as when about to use failedcmd, it's necessary to initialise $? to -1. Fix the three call sites where system might be followed by failedcmd but this wasn't done.
* Subprocess error handling: Fix when divert check failsIan Jackson2016-07-17
| | | | | If the remote command failed, do not try to match the undef value against the various regexps.
* Subprocess error handling: Do not call faildcmd when open failsIan Jackson2016-07-17
| | | | | failecmd wants to look at $? too and anyway the full command is not of much interest in these cases.
* Split brain: Fix grammar in a messageIan Jackson2016-07-17
* Split brain: In dopush, get HEAD earlier (nfc)Ian Jackson2016-07-17
* Split brain: Do not look at all of %INC for cache keyIan Jackson2016-07-17
| | | | | | | | Look only at dgit and Debian/Dgit. The prompt for this is that some of our submodules seem to dynamically load some of their components depending on what dgit does, making the cache key unstable.
* quilt innards: Change calling convention for quilt_check_splitbrain_cacheIan Jackson2016-07-17
* quilt innards: Break out quilt_check_splitbrain_cache (nfc)Ian Jackson2016-07-17
* quilt innards: Break out uilt_make_fake_dsc (nfc)Ian Jackson2016-07-17
* quilt innards: Remove a couple of stray blank linesIan Jackson2016-07-17
* quilt innards: Fix a formatting glitchIan Jackson2016-07-17
* Quilt mode reporting: Only print warning about quilt onceIan Jackson2016-07-17
* When synthesing a commit from a .dsc from the archive, stop internal git ↵Ian Jackson2016-07-16
| | | | reset from printing a confusing message about HEAD.
* Split brain: Use GZIP -1n to avoid spurious timestampIan Jackson2016-07-16
| | | | | | Otherwise the cache entry's .dsc contains the hash of a fake debian.tar.gz which has had a timestamp put in it by gzip, defeating the cache.
* Split brain: Improve and rationalise progress messagesIan Jackson2016-07-16
* Split brain: Provide --gbp= and --gbp:Ian Jackson2016-07-16
| | | | Not documented yet.
* Split brain: With patches-unapplied trees, apply before running rulesIan Jackson2016-07-16
| | | | | | | | | | | | Everywhere we invoke dpkg-buildpackage, check if we need to apply the patches. And at the end of every command that might have done this, check if we did, and unapply them if so. We don't try to unapply patches in case of error. That seems likely to be quite fragile. Print a warning when we apply patches, suggesting ways to avoid the need.
* Split brain: Show progress message when dgit view created OKIan Jackson2016-07-16
* Split brain: Do not leave STDOUT redirected to /dev/null (!)Ian Jackson2016-07-16
| | | | | We use runcmd shell_cmd for gbp pq import, and the other commands not produce undesirable output.
* Split brain: Use `gbp pq' (which exists) not `gbp-pq' (which doesn't)Ian Jackson2016-07-16
* Split brain: Include quilt mode in cache keyIan Jackson2016-07-16
| | | | This can make a radical difference!
* Split brain: Fix unapplied tree handlingIan Jackson2016-07-16
| | | | | | | Do the quiltify_splitbrain_needed if there _are_ patches, not if there aren't. And if we're told it was an unapplied tree, check that HEAD and orig are indeed the same (excluding debian/*, of course) and print a useful message if not.
* Split build: Add a bit more debugging outputIan Jackson2016-07-16
* Split brain: Run source build in dgit view directoryIan Jackson2016-07-16
| | | | | And, after the source build is done, copy the resulting files out into the invocation directory's parent, where people expect to find them.
* Split build: Split clean from source build in split build modeIan Jackson2016-07-16
* Split brain: some work on integration into the rest of dgitIan Jackson2016-07-16
| | | | | | | | | | | * Move the fast forward die to where we will need to make the synthetic commit, so that the tree equality check is done on the thing we will actually push. * Crash if we are trying to do split brain without split build. (This should never happen.) * Set split build if we need split brain.
* dgit: Rename $suppress_clean to $clean_using_builder (nfc)Ian Jackson2016-07-16
| | | | | This variable name was very confusing. Also, add a comment explaining the semantics.
* Split brain: Quieten dgit-view checkoutIan Jackson2016-07-16
* Split brain: introduce quiltmode_splitbrain() (nfc)Ian Jackson2016-07-16
* Split brain: Memoise (cache) split brain dgit-view in a reflogIan Jackson2016-07-16
* Split brain: Actually get gbp pq output back onto dgit-view branchIan Jackson2016-07-16
| | | | And explain why in a comment.
* Split brain: Transfer gbp-pq patch application output to dgit-view branchIan Jackson2016-07-16
| | | | This code does not work properly yet; it will be fixed soon.