templater: resurrect cache of engine instance The engine-level cache was effectively disabled at 48289eafb37d "templater: drop extension point of engine classes (API)" by mistake, which made template rendering quite slow. Spotted by Martin von Zweigbergk.

templater: extract template loader to separate class This avoids reference cycle in the subsequent patch: templater -> _proc -> templater.load -> templater The templater class will be a thin wrapper around the loader and the engine.

tests: rename and document test-command-template.t So that we can specify templater tests as test-template-*. (original) $ wc -l test-command-template.t test-template-* 5041 test-command-template.t 23 test-template-filters.t 5064 total (at this patch) $ wc -l test-template-* 1088 test-template-basic.t 1376 test-template-functions.t 1195 test-template-keywords.t 1760 test-template-map.t 5419 total

tests: extract test-template-keywords.t from test-command-template.t

tests: fold test-template-filters.t into test-template-functions.t

tests: extract test-template-functions.t from test-command-template.t I decided to not split filters and functions into two test files since we sometimes reimplement a filter as a function.

tests: extract test-template-map.t from test-command-template.t test-command-template.t is one of the slowest tests. Let's split it into 4 files of manageable size.

cleanupnodes: preserve phase of parents of new nodes As Yuya noted in the review of D3818, passing in targetphase=phases.draft would result in advancing the phase boundary of a secret-phase parent. We never pass targetphase=phases.draft so far, but it's a bug waiting to happen. I tried to refactor it so max(parentphase, X) happened in one place, but I couldn't come up with good variables names and I ended up with a "newphase = max(newphase, parentphase)" line, which made the whole block not look any better to me. Differential Revision: https://phab.mercurial-scm.org/D3824

rebase: delete the comment which was not following "do not eat my data" Differential Revision: https://phab.mercurial-scm.org/D3756

tags: unblock log-like template keywords and functions It checks if ctx will be used in template since loading ctx per revision could take extra 10-100msec in total depending on the number of tags.

formatter: provide hint of context keys required by template This allows us to create ctx objects only if necessary. I tried another idea which is to populate ctx from 'repo' and 'node' value on demand. It worked, but seemed unnecessarily complicated. So I chose a simpler one. The datafields argument is a space-separated string for consistency with fm.write() API.

templatefuncs: declare resource requirements for future use

templatefuncs: minimize resource requirements

merge with stable

merge: do not fill manifest of committed revision with pseudo node (issue5526) Since a75d24539aba "convert: fix convert dropping p2 contents during filemap merge", wctx is not always a committablectx because the convert extension passes in repo[n] as wctx. If wctx is a committed changeset, its manifest dict shouldn't be mutated reflecting to the working directory.

split: preserve phase of commit that is being split With this change, hg split will preserve the phase of the commit that is being split, ignoring the phases.new-commit setting. Previously, we would use whatever phases.new-commit was set to (unless our parent was secret, then we would be secret even if phases.new-commit=draft). Now, splitting a draft commit with phases.new-commit=secret does not cause the new commits to become secret, and splitting a secret commit with phases.new-commit=draft and a draft parent does not cause the new commits to become draft. Test cases and commit message taken from Kyle Lippincott's D2016 (thanks!). Differential Revision: https://phab.mercurial-scm.org/D3819

scmutil: make cleanupnodes optionally also fix the phase We have had multiple bugs where the phase wasn't correctly carried forward to a rewritten changeset (for example: phabricator, split, evolve, fix). Handling the phase update in cleanupnodes() makes it less likely to happen again, especially once we have made it fix the phase by default (perhaps in the next release cycle). This patch also updates all applicable callers so we get some testing of it. Note that rebase and histedit can't be fixed yet because they call cleanupnodes() only at the end and the phase may have been changed by the user when the rebase/histedit was interrupted (due to merge conflicts). I think we should make them write one commit at a time (as it already does), along with associated obsmarkers, bookmark moves, etc. When that's done, we can switch them over to cleanupnodes(). Differential Revision: https://phab.mercurial-scm.org/D3818

tests: add test of uncommit with default phase as secret We didn't seem to have any test checking that uncommitting a draft commit with phase.new-phase=secret preserved the draft phase. Differential Revision: https://phab.mercurial-scm.org/D3817

grep: add --diff flag Adds a diff flag, which works exactly same as all, in fact since --all searches diffs, there diff is a better name for it. The all flag is still here for backward compatibility reasons. Some major tests for all has been picked and added for diff. Differential Revision: https://phab.mercurial-scm.org/D3763

fix: include cleanupnodes() in transaction As pointed out by Yuya, we need a transaction to make sure the state before the call to cleanupnodes() is not observable. Differential Revision: https://phab.mercurial-scm.org/D3823

progress: enforce use of complete() on the helper class complete() is preferred over update(None), so let's enforce that. Differential Revision: https://phab.mercurial-scm.org/D3822

progress: extract function for closing topic progress(None) had a completely different implementation from the progress(<not None>) implementation. It very much feels like it should be a separate method, so this patch makes it so. That also makes it clear that only the topic parameter matters when closing a topic (e.g. "total" does not matter). Differential Revision: https://phab.mercurial-scm.org/D3821

progress: use context manager for lock Differential Revision: https://phab.mercurial-scm.org/D3820

configitem: reorder items in the 'server' section Keeping things alphabetically sorted.

test-lfs: add coverage for the binary() fileset This ensures that the blobs don't need to be present to be filtered properly.

fileset: use filectx.isbinary() to filter out binaries in eol() Since LFS stores the binary attribute in the pointer file, this means that the file doesn't need to be downloaded in order to be skipped. This function also catches an IOError if the data can't be loaded in the non-LFS case. I wonder if it's worth storing the unix/dos attributes in the pointer file as well, though I'd expect LFS files to be binary most of the time.

crecord: re-center display in interactive curses commit on pageup/down A long-standing issue in the crecord (interactive curses commit interface) is that using PageUp/Down to move along longer-than current screen size chunks would "lose" the cursor and not properly re-center. There has been self.recenterdisplayedarea() to do that, but it has not been in use for some reason. Add calls to the appropriate uparrowshiftevent() and downarrowshiftevent() methods to fix this.

tests: in test-split.t, save a "clean" copy of pre-split repo for later use Differential Revision: https://phab.mercurial-scm.org/D2015

tests: use progress helper With this commit, the only in-tree caller of ui.progress() is scmutil.progress(). That means that we could deprecate it. It also means that we can considering inlining it in scmutil.progress. Differential Revision: https://phab.mercurial-scm.org/D3812

changegroup: use progress helper Although it looks like this code was micro-optimized, I could not measure any slow-down. Differential Revision: https://phab.mercurial-scm.org/D3811