git-debrebase: Set $|

This will avoid pepeering everything with calls to flush.  Without it
some messages from ffq_check, say, can arrive after the stderr output
from a subsequent snags_maybe_bail, which is quite confusing.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

git-debrebase: Provide snags_maybe_bail_early

No callers yet.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

git-debrebase: Replace `die badusage' with `badusage'.

badusage contains a call to finish.  So the die is redundant and never
gets called.

No functional change.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

git-debrebase: ffq_check: Move defaults for $ff and $notff into sub

No functional change.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

Dgit.pm: Move debinatag_maintview from dgit

git-debrebase is about to want this for convert-from-gbp dgit split
brain fixup.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

Dgit.pm: Break out parsechangelog_loop from dgit

git-debrebase is going to want this.

No functional change except to some debugging messages.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

Dgit.pm: Move parsecontrol etc. out of dgit

No functional change.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

dgit: Fix a typo in a comment

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

git-debrebase: Break out ff_check (for reuse by convert-from-gbp)

No functional change.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

git-debrebase: record_ffq_prev_deferred: Move get_head();

No functional change.  This will make ff check easier to split out.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

changelog: start 5.9

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

changelog: finalise 5.8

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

git-debrebase: Improve grammar if one blocking snag.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

test suite: gdr-newupstream: Test adding revision

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

git-debrebase new-upstream: Add a -1 revision if the user didn't supply one.

Closes:#903127.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

git-debrebase: status: Reorganise commit reporting.

Replace the references to @_ with a hash.  Using a hash is much less
confusing than all this numerical indexing.  Also, \@_ is wrong
because there is only one @_, so by the time it is dereference, the
commit being printed is the anchor.

This is why in #903131 we see

      branch is unlaundered
      found mixed upstream/packaging commit (7badba627162337c7057002f37e9a1a593d07d08)
        abcc6260 Update to upstream 2.9.7

which appears to sugggest that the anchor is a mixed commit.  Actually
the mixed commit is 7badba627162 but the subsequent message is based
on @_ which has different information in it by then.

And, set $kcmsg from the main message, not including the commit
information.  We are the more sophisticated caller mentioned in the
previous commmit.

Closes: #903131.
Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

git-debrebase: keycommits: Callbacks get separate $mainwhy

Separate out the extra contextual info, naming the hash of the commit
being complained about, into a separate argument to $x->().  This
allows us to pass the message without that as an additional argument
to the callbacks.

This way we can still pass \&snag as a callback but other callers can
do something more sophisticated involving $cl.

No functional change for any existing callers.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

test suite: makepatches7: Add a nasty .gitignore

Prior to the fix for #903130 this prevents both dgit and gdr from
making patches properly.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

dgit, git-debrebase: Properly make patches with nasty .gitignores.

Specifically:

 * When running git status --porcelain, pass --ignored so it
   lists ignored files, and handle the ! output that it then
   produces.

 * Add missing -f to a few git add invocations.

I have done some greps ('git.*add', 'git.*status', 'porcelain', etc.)
to try to find other missed cases and none turned up.

git diff is OK because normally we pass two treeish arguments, in
which case the ignores are ignored by git diff.  When we are asking it
to look at the working tree, we are expecting it to ignore untracked
files (whether ignored or not), and diff tracked ones, which is what
it does.

Closes:#903130.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

changelog: Document dgit(1) build unscrambling

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
squash! changelog: Document manpage fix

dgit(1): Unscramble push and push-source descriptions

These got mixed up in a recent commit.

Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>

changelog: start 5.8

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

changelog: finalise 5.7

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

test suite: sourceonlypolicy: New test

Test the fix for #801435.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

dgit: Check that entirely-new uploads to Debian are not source-only-uploads.

Avoids the user pushing things that will inevitably be REJECTed.

We end up introducing some new config and force options to support
this.

We do not reuse test_source_only_changes, as it doesn't have quite the
right shape (in particular, it sometimes blithers on stdout).  Also
arguably it is better to look, specifically, for .debs, for the
purpose of this test.

This requires a new archive protocol method.  We implement it for
ftpmasterapi (where it is actualliy needed) and dummycatapi (where we
need it for tests.)  Implementing it for madisonish methods would be
easy.  For aptget, it would probably involve iterating over suites,
and not be adviseable.

Closes:#801435.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

test suite: Test dput-ng compatibility.

Add dput-ng as a dependency to trustingpolicy-replay.

This is the test which went wrong with dput-ng before.  It also has a
number of pushes, so it will do.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

test suite: Fix trustingpolicy-replay & dput-ng.

Specifically:

 * Only dput-ng actually complains about the existence of .upload
   files.  dput itself just says there is nothing to do and calls that
   success.

 * The call to t-rm-dput-dropping was in the wrong place.  t-commit
   bumps the version to 1.1 but it's 1.0 that needs removing.

Closes:#903007.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

test suite: checkout: Check we end up on the right branch each time

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

dgit: checkout: Do not leave detached HEAD!

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

test suite: Always pass LC_COLLATE=C to sort(1).

Specifically, by replacing every invocation of sort with the new
function t-sort.  The one case where we already took care of this, we
change LC_ALL to LC_COLLATE, which should be sufficient.

Closes:#903006.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

dgit(7): Mention git-debrebase and gbp pq alongside git-dpm

in the comment about handling patch stacks.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

test suite: checkout: new test

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

dgit(1): checkout: Document when this acceses the network

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

dgit: checkout: New command.

Closes:#878443.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

dgit: fetch: Introduce new dofetch()

This actually does all that is needed.  It replaces two now-identical
calling patterns.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

dgit: fetch: Reorganise cmd_fetch

This slightly odd construction is actually how the return value from
fork_for_multisuite should be handled.  It now looks much more like
the calling pattern in pull().

No functional change: if $multi_fetched is falseish, we call
fetch_one, and then, no matter what, we finish 0.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

dgit: fetch: Slightly refactor pull()

When for_for_multisuite returns '0', meaning we are a child and should
do one suite, use finish 0 rather than returning, after doing the
fetch.

This has no functional change because the one call site is at the end
of cmd_pull, after which we finish 0 anyway.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

dgit: fetch: Rename fetch_one from fetch (nfc)

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

dgit(1): Document as-yet-unwritten `dgit checkout'

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

dgit: update-vcs-git: Honour --package properly.

Specifically, use package_from_d_control rather than open-coding
a substandard implementation.

The consequent abolition of $sourcep does mean that if the user
specifies the suite `.', and does not specify -p, we needlessly parse
debian/control twice.  This doesn't really matter.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

dgit: Break out package_from_d_control (nfc)

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

changelog: start 5.7

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

changelog: finalise 5.6

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

changelog: Document empirical effectiveness of new gnupg workaround

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

test suite: manpages-format: Test -Tps too

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

test suite: manpages-format: work properly when no expected warnings

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

test suite: manpages-format: Remove some pointless redirects

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

test suite: manpages-format: Prep for other formats (nfc)

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

test suite: manpages-format: New test

Test that manpages format with only expected roff warnings.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

Merge remote-tracking branch 'dgit/dgit/experimental'

Conflicts:
debian/changelog

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

test suite: gpg agent workaround: Fix bugs in log neatening.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

test suite: gpg agent workaround: Neaten the logging.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

test suite: Provide DGIT_TEST_BYHAND if appropriate

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

test suite: gnupg workaround: Remove a sprurious $tmp

This is not set here, which is just as well because it would have been
quite wrong.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

test suite: gpg-agent workaround: Send ps output to stderr (oops)

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

test suite: gpg-agent workaround: Cope if ps not found.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

changelog: finalise 5.5+exp5

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

test suite: gpg-agent workaround: Kill unexpected leftover agents.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

test suite: gpg-agent workaround: Add more debugging output.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

test suite: Add bc to appropriate test Depends.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

changelog: 5.5+exp3 to placate version checking

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

Merge remote-tracking branch 'dgit/dgit/sid' into experimental

changelog: start 5.6

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

changelog: finalise 5.5

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

dgit(1): Fix a wrong reference to \fp, which should be \fP.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

debian/control: Add missing comma

Closes:#902578.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

New gnupg agent thing

before (5.3)

1044.27user 126.77system 6:26.89elapsed 302%CPU (0avgtext+0avgdata 402420maxresident)k
1778718inputs+5123720outputs (2370major+57693524minor)pagefaults 0swaps

after (5.3 + these changes)

1072.81user 131.38system 6:25.02elapsed 312%CPU (0avgtext+0avgdata 402448maxresident)k
1777558inputs+5124952outputs (2370major+59176578minor)pagefaults 0swaps

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

No-change upload, for testing.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

Merge remote-tracking branch 'dgit/dgit/experimental' into HEAD

-s ours

changelog: start 5.5

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

changelog: finalise 5.4

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

No-change upload, for testing.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

test suite: gdr-viagit, gdr-newupstream: do not auto-import t-setup-gnupg.

We provide a new mechanism for individual tests to disable the
automatic t-setup-import of gnupg.

We need to add this to all the other setup/ that these import, too.
gdr-convert-gbp is particularly awkward because it wants gnupg if
$ifarchive isn't `:'.  There we arrange to suppress its automatic
inclusion, but add it again by hand if appropriate.

This fixes these two tests in chroots without gnupg.

Closes:#902559.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

dgit(1): Better description of --overwrite.

In particular, be clear that --overwrite (without previous-version) is
quite a weak promise: that the version you are uploading contains
everything in your changelog.  It won't overwrite willy-nilly.

Somewhat apropos of discussion in #902534.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

changelog: Fix bug ref to #865444 in previous changelog entry.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

changelog: start 5.4

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

changelog: finalise 5.3

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

dgit(7): Add discussion of quilt fixup error messages

And add cross-references to dgit(1) and the actual error.
Somewhat apropos of #842614.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

dgit: Better message formatting when --overwrite may be needed

Closes:891031.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

test suite: --overwrite test: Be more tolerant of exact message

We're about to change this, apropos of #891031.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

dgit(1): Mention in --overwrite that it is usually needed for first dgit push

This is part of #891031.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

test suite: Test ill-advised repeat dgit push after git push fails

This demonstrates #869146, which is about dgit repeatedly adding the
same .orig to the .changes.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

dgit: Do not introduce duplicate origs in .changes files

The .dsc and .changes matching code contained roughly
   $fieldbody =~ / \Q$item\E\n/m
but this is wrong when the $fieldbody matches $item in its last line,
and there is no trailing newline (as would be usual).  This code
is obviouslu confused because m affects only ^ $ which don't appear
in the regexp.  Change these to / \Q$item\E$/m which is what ought
to have been, and probably was, intended.

In one case we are trying to delete a line.  In that case we need to
remove the leading newline.  The field body always starts with a
newline so the use of \n rather than ^ is fine.

Closes:#869146.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

test suite: Test GIT_REFLOG_ACTION setting by git-debrebase

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

git-debrebase new-upstream: Provide better reflog entries

By setting GIT_REFLOG_ACTION.

Closes:#901925.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

ref updates: Break out git_reflog_action_msg

No functional change.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

ref updates: Honour GIT_REFLOG_ACTION everywhere.

Specifically, every time we call git_update_ref_cmd.

Closes:#901935.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

ref updates: Introduce update_ref_cmd

Currently the comment about `message may be modified' is rather
misleading, since there is no code to do that - but it explains the
semantics of the new function.

We change the appropriate subset of the git update-ref call sites.
These are precisely the ones where we previously specified a message
with -m, except for the quilt cache (where we use the reflog message
as the cache lookup key).

The calls are updates of internal refs etc.

We end up using qw(git) rather than @git.
Otherwise no functional change.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

changelog: start 5.3~

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

changelog: finalise 5.2

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

test suite: dgit/gdr interop: test make-patches exiting 7

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

test suite: lib-gdr: t-some-changes: Honour $2 `dum'

This allows for less thorough but faster test cases.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

dgit: git-debrebase interop: Actually tolerate git-debrebase status 7

make-patches exiting with status 7 causes $?==7*256, not 7.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

dgit: git-debrebase interop: Add a missing debugcmd debugging print.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

dgit: warn if vcs-git seems out of date

Print a warning to stderr on `dgit fetch sid', if your vcs-git remote
url disagrees with what's in sid's .dsc.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

dgit: Provide `update-vcs-git' subcommand

This is for creating and adjusting the vcs-git remote url.  Useful for
transition from alioth to salsa.  Closes:#902006.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

dgit-user(7): Suggest sbuild-debian-developer-setup(1)

This will only create the chroot for building packages for the
machine's architecture, because sbuild-debian-developer-setup(1) does
not have an --architecture option, so sbuild-createchroot(1) will be
needed too, hence the 'and'.

The nice thing about sbuild-debian-developer-setup(1) is that does
fiddly things that a non-expert might forget about, like adding the
current user to the sbuild group.  So it is still worth recommending
even though it does not have an --architecture option.

Closes:#895779.

Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>

test suite: Test that --new --overwrite works properly.

Here --overwrite is not needed, so is a no-op.  I chose this test for
no particular reason, out of the results of git-grep -e --new.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

dgit(1): Clarify that --overwrite does nothing if not needed.

Closes:#863578.  Related to #863576 too.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

dgit: --overwrite now no longer crashes if there is no $archive_hash

Ie there is nothing to overwrite (eg, when used with --new).  Instead,
it is simply ignored, as it is ignored in situations where the push is
fast forward.

Closes:#863576.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>