narrow: fix commits of empty files

The problem is that when committing a new file with empty contents (or
in general empty file with filelog p1 = -1), hg commit with narrow
doesn't create a filelog revision at all, which causes failures in
further commands.

The problem seems to be that:
- hg thinks that instead of creating a new filelog revision, it can
  use the filelog's p1 (the nullrev)
- because it thinks the file contents is the same in that revision and
  in p1
- because `narrowfilelog.cmp(nullrev, b'')` is True (unlike with
  `filelog.cmp`)

It's not clear to me which `cmp` behaves better. But I think it makes
sense to change the commit code to not to "reuse" the null rev when
adding an empty file with filelog p1 == filelog p2 == -1. This is
consistent with never writing the null rev in the manifest, which `hg
verify` claims is an invariant:

```
inside/c@4: manifest refers to unknown revision 000000000000
```

Differential Revision: https://phab.mercurial-scm.org/D11400

mercurial/commit.py

@@ -1,503 +1,504 b''
 # commit.py - fonction to perform commit
 #
 # This software may be used and distributed according to the terms of the
 # GNU General Public License version 2 or any later version.
 
 from __future__ import absolute_import
 
 import errno
 
 from .i18n import _
 from .node import (
     hex,
     nullrev,
 )
 
 from . import (
     context,
     mergestate,
     metadata,
     phases,
     scmutil,
     subrepoutil,
 )
 
 
 def _write_copy_meta(repo):
     """return a (changelog, filelog) boolean tuple
 
     changelog: copy related information should be stored in the changeset
     filelof:   copy related information should be written in the file revision
     """
     if repo.filecopiesmode == b'changeset-sidedata':
         writechangesetcopy = True
         writefilecopymeta = True
     else:
         writecopiesto = repo.ui.config(b'experimental', b'copies.write-to')
         writefilecopymeta = writecopiesto != b'changeset-only'
         writechangesetcopy = writecopiesto in (
             b'changeset-only',
             b'compatibility',
         )
     return writechangesetcopy, writefilecopymeta
 
 
 def commitctx(repo, ctx, error=False, origctx=None):
     """Add a new revision to the target repository.
     Revision information is passed via the context argument.
 
     ctx.files() should list all files involved in this commit, i.e.
     modified/added/removed files. On merge, it may be wider than the
     ctx.files() to be committed, since any file nodes derived directly
     from p1 or p2 are excluded from the committed ctx.files().
 
     origctx is for convert to work around the problem that bug
     fixes to the files list in changesets change hashes. For
     convert to be the identity, it can pass an origctx and this
     function will use the same files list when it makes sense to
     do so.
     """
     repo = repo.unfiltered()
 
     p1, p2 = ctx.p1(), ctx.p2()
     user = ctx.user()
 
     with repo.lock(), repo.transaction(b"commit") as tr:
         mn, files = _prepare_files(tr, ctx, error=error, origctx=origctx)
 
         extra = ctx.extra().copy()
 
         if extra is not None:
             for name in (
                 b'p1copies',
                 b'p2copies',
                 b'filesadded',
                 b'filesremoved',
             ):
                 extra.pop(name, None)
         if repo.changelog._copiesstorage == b'extra':
             extra = _extra_with_copies(repo, extra, files)
 
         # save the tip to check whether we actually committed anything
         oldtip = repo.changelog.tiprev()
 
         # update changelog
         repo.ui.note(_(b"committing changelog\n"))
         repo.changelog.delayupdate(tr)
         n = repo.changelog.add(
             mn,
             files,
             ctx.description(),
             tr,
             p1.node(),
             p2.node(),
             user,
             ctx.date(),
             extra,
         )
         rev = repo[n].rev()
         if oldtip != repo.changelog.tiprev():
             repo.register_changeset(rev, repo.changelog.changelogrevision(rev))
 
         xp1, xp2 = p1.hex(), p2 and p2.hex() or b''
         repo.hook(
             b'pretxncommit',
             throw=True,
             node=hex(n),
             parent1=xp1,
             parent2=xp2,
         )
         # set the new commit is proper phase
         targetphase = subrepoutil.newcommitphase(repo.ui, ctx)
 
         # prevent unmarking changesets as public on recommit
         waspublic = oldtip == repo.changelog.tiprev() and not repo[rev].phase()
 
         if targetphase and not waspublic:
             # retract boundary do not alter parent changeset.
             # if a parent have higher the resulting phase will
             # be compliant anyway
             #
             # if minimal phase was 0 we don't need to retract anything
             phases.registernew(repo, tr, targetphase, [rev])
         return n
 
 
 def _prepare_files(tr, ctx, error=False, origctx=None):
     repo = ctx.repo()
     p1 = ctx.p1()
 
     writechangesetcopy, writefilecopymeta = _write_copy_meta(repo)
     files = metadata.ChangingFiles()
     ms = mergestate.mergestate.read(repo)
     salvaged = _get_salvaged(repo, ms, ctx)
     for s in salvaged:
         files.mark_salvaged(s)
 
     if ctx.manifestnode():
         # reuse an existing manifest revision
         repo.ui.debug(b'reusing known manifest\n')
         mn = ctx.manifestnode()
         files.update_touched(ctx.files())
         if writechangesetcopy:
             files.update_added(ctx.filesadded())
             files.update_removed(ctx.filesremoved())
     elif not ctx.files():
         repo.ui.debug(b'reusing manifest from p1 (no file change)\n')
         mn = p1.manifestnode()
     else:
         mn = _process_files(tr, ctx, ms, files, error=error)
 
     if origctx and origctx.manifestnode() == mn:
         origfiles = origctx.files()
         assert files.touched.issubset(origfiles)
         files.update_touched(origfiles)
 
     if writechangesetcopy:
         files.update_copies_from_p1(ctx.p1copies())
         files.update_copies_from_p2(ctx.p2copies())
 
     return mn, files
 
 
 def _get_salvaged(repo, ms, ctx):
     """returns a list of salvaged files
 
     returns empty list if config option which process salvaged files are
     not enabled"""
     salvaged = []
     copy_sd = repo.filecopiesmode == b'changeset-sidedata'
     if copy_sd and len(ctx.parents()) > 1:
         if ms.active():
             for fname in sorted(ms.allextras().keys()):
                 might_removed = ms.extras(fname).get(b'merge-removal-candidate')
                 if might_removed == b'yes':
                     if fname in ctx:
                         salvaged.append(fname)
     return salvaged
 
 
 def _process_files(tr, ctx, ms, files, error=False):
     repo = ctx.repo()
     p1 = ctx.p1()
     p2 = ctx.p2()
 
     writechangesetcopy, writefilecopymeta = _write_copy_meta(repo)
 
     m1ctx = p1.manifestctx()
     m2ctx = p2.manifestctx()
     mctx = m1ctx.copy()
 
     m = mctx.read()
     m1 = m1ctx.read()
     m2 = m2ctx.read()
 
     # check in files
     added = []
     removed = list(ctx.removed())
     linkrev = len(repo)
     repo.ui.note(_(b"committing files:\n"))
     uipathfn = scmutil.getuipathfn(repo)
     for f in sorted(ctx.modified() + ctx.added()):
         repo.ui.note(uipathfn(f) + b"\n")
         try:
             fctx = ctx[f]
             if fctx is None:
                 removed.append(f)
             else:
                 added.append(f)
                 m[f], is_touched = _filecommit(
                     repo, fctx, m1, m2, linkrev, tr, writefilecopymeta, ms
                 )
                 if is_touched:
                     if is_touched == 'added':
                         files.mark_added(f)
                     elif is_touched == 'merged':
                         files.mark_merged(f)
                     else:
                         files.mark_touched(f)
                 m.setflag(f, fctx.flags())
         except OSError:
             repo.ui.warn(_(b"trouble committing %s!\n") % uipathfn(f))
             raise
         except IOError as inst:
             errcode = getattr(inst, 'errno', errno.ENOENT)
             if error or errcode and errcode != errno.ENOENT:
                 repo.ui.warn(_(b"trouble committing %s!\n") % uipathfn(f))
             raise
 
     # update manifest
     removed = [f for f in removed if f in m1 or f in m2]
     drop = sorted([f for f in removed if f in m])
     for f in drop:
         del m[f]
     if p2.rev() == nullrev:
         files.update_removed(removed)
     else:
         rf = metadata.get_removal_filter(ctx, (p1, p2, m1, m2))
         for f in removed:
             if not rf(f):
                 files.mark_removed(f)
 
     mn = _commit_manifest(tr, linkrev, ctx, mctx, m, files.touched, added, drop)
 
     return mn
 
 
 def _filecommit(
     repo,
     fctx,
     manifest1,
     manifest2,
     linkrev,
     tr,
     includecopymeta,
     ms,
 ):
     """
     commit an individual file as part of a larger transaction
 
     input:
 
         fctx:       a file context with the content we are trying to commit
         manifest1:  manifest of changeset first parent
         manifest2:  manifest of changeset second parent
         linkrev:    revision number of the changeset being created
         tr:         current transation
         includecopymeta: boolean, set to False to skip storing the copy data
                     (only used by the Google specific feature of using
                     changeset extra as copy source of truth).
         ms:         mergestate object
 
     output: (filenode, touched)
 
         filenode: the filenode that should be used by this changeset
         touched:  one of: None (mean untouched), 'added' or 'modified'
     """
 
     fname = fctx.path()
     fparent1 = manifest1.get(fname, repo.nullid)
     fparent2 = manifest2.get(fname, repo.nullid)
     touched = None
     if fparent1 == fparent2 == repo.nullid:
         touched = 'added'
 
     if isinstance(fctx, context.filectx):
         # This block fast path most comparisons which are usually done. It
         # assumes that bare filectx is used and no merge happened, hence no
         # need to create a new file revision in this case.
         node = fctx.filenode()
         if node in [fparent1, fparent2]:
             repo.ui.debug(b'reusing %s filelog entry\n' % fname)
             if (
                 fparent1 != repo.nullid
                 and manifest1.flags(fname) != fctx.flags()
             ) or (
                 fparent2 != repo.nullid
                 and manifest2.flags(fname) != fctx.flags()
             ):
                 touched = 'modified'
             return node, touched
 
     flog = repo.file(fname)
     meta = {}
     cfname = fctx.copysource()
     fnode = None
 
     if cfname and cfname != fname:
         # Mark the new revision of this file as a copy of another
         # file.  This copy data will effectively act as a parent
         # of this new revision.  If this is a merge, the first
         # parent will be the nullid (meaning "look up the copy data")
         # and the second one will be the other parent.  For example:
         #
         # 0 --- 1 --- 3   rev1 changes file foo
         #   \       /     rev2 renames foo to bar and changes it
         #    \- 2 -/      rev3 should have bar with all changes and
         #                      should record that bar descends from
         #                      bar in rev2 and foo in rev1
         #
         # this allows this merge to succeed:
         #
         # 0 --- 1 --- 3   rev4 reverts the content change from rev2
         #   \       /     merging rev3 and rev4 should use bar@rev2
         #    \- 2 --- 4        as the merge base
         #
 
         cnode = manifest1.get(cfname)
         newfparent = fparent2
 
         if manifest2:  # branch merge
             if (
                 fparent2 == repo.nullid or cnode is None
             ):  # copied on remote side
                 if cfname in manifest2:
                     cnode = manifest2[cfname]
                     newfparent = fparent1
 
         # Here, we used to search backwards through history to try to find
         # where the file copy came from if the source of a copy was not in
         # the parent directory. However, this doesn't actually make sense to
         # do (what does a copy from something not in your working copy even
         # mean?) and it causes bugs (eg, issue4476). Instead, we will warn
         # the user that copy information was dropped, so if they didn't
         # expect this outcome it can be fixed, but this is the correct
         # behavior in this circumstance.
 
         if cnode:
             repo.ui.debug(b" %s: copy %s:%s\n" % (fname, cfname, hex(cnode)))
             if includecopymeta:
                 meta[b"copy"] = cfname
                 meta[b"copyrev"] = hex(cnode)
             fparent1, fparent2 = repo.nullid, newfparent
         else:
             repo.ui.warn(
                 _(
                     b"warning: can't find ancestor for '%s' "
                     b"copied from '%s'!\n"
                 )
                 % (fname, cfname)
             )
 
     elif fparent1 == repo.nullid:
         fparent1, fparent2 = fparent2, repo.nullid
     elif fparent2 != repo.nullid:
         if ms.active() and ms.extras(fname).get(b'filenode-source') == b'other':
             fparent1, fparent2 = fparent2, repo.nullid
         elif ms.active() and ms.extras(fname).get(b'merged') != b'yes':
             fparent1, fparent2 = fparent1, repo.nullid
         # is one parent an ancestor of the other?
         else:
             fparentancestors = flog.commonancestorsheads(fparent1, fparent2)
             if fparent1 in fparentancestors:
                 fparent1, fparent2 = fparent2, repo.nullid
             elif fparent2 in fparentancestors:
                 fparent2 = repo.nullid
 
     force_new_node = False
     # The file might have been deleted by merge code and user explicitly choose
     # to revert the file and keep it. The other case can be where there is
     # change-delete or delete-change conflict and user explicitly choose to keep
     # the file. The goal is to create a new filenode for users explicit choices
     if (
         repo.ui.configbool(b'experimental', b'merge-track-salvaged')
         and ms.active()
         and ms.extras(fname).get(b'merge-removal-candidate') == b'yes'
     ):
         force_new_node = True
     # is the file changed?
     text = fctx.data()
     if (
         fparent2 != repo.nullid
+        or fparent1 == repo.nullid
         or meta
         or flog.cmp(fparent1, text)
         or force_new_node
     ):
         if touched is None:  # do not overwrite added
             if fparent2 == repo.nullid:
                 touched = 'modified'
             else:
                 touched = 'merged'
         fnode = flog.add(text, meta, tr, linkrev, fparent1, fparent2)
     # are just the flags changed during merge?
     elif fname in manifest1 and manifest1.flags(fname) != fctx.flags():
         touched = 'modified'
         fnode = fparent1
     else:
         fnode = fparent1
     return fnode, touched
 
 
 def _commit_manifest(tr, linkrev, ctx, mctx, manifest, files, added, drop):
     """make a new manifest entry (or reuse a new one)
 
     given an initialised manifest context and precomputed list of
     - files: files affected by the commit
     - added: new entries in the manifest
     - drop:  entries present in parents but absent of this one
 
     Create a new manifest revision, reuse existing ones if possible.
 
     Return the nodeid of the manifest revision.
     """
     repo = ctx.repo()
 
     md = None
 
     # all this is cached, so it is find to get them all from the ctx.
     p1 = ctx.p1()
     p2 = ctx.p2()
     m1ctx = p1.manifestctx()
 
     m1 = m1ctx.read()
 
     if not files:
         # if no "files" actually changed in terms of the changelog,
         # try hard to detect unmodified manifest entry so that the
         # exact same commit can be reproduced later on convert.
         md = m1.diff(manifest, scmutil.matchfiles(repo, ctx.files()))
     if not files and md:
         repo.ui.debug(
             b'not reusing manifest (no file change in '
             b'changelog, but manifest differs)\n'
         )
     if files or md:
         repo.ui.note(_(b"committing manifest\n"))
         # we're using narrowmatch here since it's already applied at
         # other stages (such as dirstate.walk), so we're already
         # ignoring things outside of narrowspec in most cases. The
         # one case where we might have files outside the narrowspec
         # at this point is merges, and we already error out in the
         # case where the merge has files outside of the narrowspec,
         # so this is safe.
         mn = mctx.write(
             tr,
             linkrev,
             p1.manifestnode(),
             p2.manifestnode(),
             added,
             drop,
             match=repo.narrowmatch(),
         )
     else:
         repo.ui.debug(
             b'reusing manifest from p1 (listed files ' b'actually unchanged)\n'
         )
         mn = p1.manifestnode()
 
     return mn
 
 
 def _extra_with_copies(repo, extra, files):
     """encode copy information into a `extra` dictionnary"""
     p1copies = files.copied_from_p1
     p2copies = files.copied_from_p2
     filesadded = files.added
     filesremoved = files.removed
     files = sorted(files.touched)
     if not _write_copy_meta(repo)[1]:
         # If writing only to changeset extras, use None to indicate that
         # no entry should be written. If writing to both, write an empty
         # entry to prevent the reader from falling back to reading
         # filelogs.
         p1copies = p1copies or None
         p2copies = p2copies or None
         filesadded = filesadded or None
         filesremoved = filesremoved or None
 
     extrasentries = p1copies, p2copies, filesadded, filesremoved
     if extra is None and any(x is not None for x in extrasentries):
         extra = {}
     if p1copies is not None:
         p1copies = metadata.encodecopies(files, p1copies)
         extra[b'p1copies'] = p1copies
     if p2copies is not None:
         p2copies = metadata.encodecopies(files, p2copies)
         extra[b'p2copies'] = p2copies
     if filesadded is not None:
         filesadded = metadata.encodefileindices(files, filesadded)
         extra[b'filesadded'] = filesadded
     if filesremoved is not None:
         filesremoved = metadata.encodefileindices(files, filesremoved)
         extra[b'filesremoved'] = filesremoved
     return extra

tests/test-narrow-commit.t

@@ -1,118 +1,111 b''
 #testcases flat tree
 
   $ . "$TESTDIR/narrow-library.sh"
 
 #if tree
   $ cat << EOF >> $HGRCPATH
   > [experimental]
   > treemanifest = 1
   > EOF
 #endif
 
 create full repo
 
   $ hg init master
   $ cd master
 
   $ mkdir inside
   $ echo inside > inside/f1
   $ mkdir outside
   $ echo outside > outside/f1
   $ hg ci -Aqm 'initial'
 
   $ echo modified > inside/f1
   $ hg ci -qm 'modify inside'
 
   $ echo modified > outside/f1
   $ hg ci -qm 'modify outside'
 
   $ cd ..
 
 (The lfs extension does nothing here, but this test ensures that its hook that
 determines whether to add the lfs requirement, respects the narrow boundaries.)
 
   $ hg --config extensions.lfs= clone --narrow ssh://user@dummy/master narrow \
   >    --include inside
   requesting all changes
   adding changesets
   adding manifests
   adding file changes
   added 3 changesets with 2 changes to 1 files
   new changesets *:* (glob)
   updating to branch default
   1 files updated, 0 files merged, 0 files removed, 0 files unresolved
   $ cd narrow
 
   $ hg update -q 0
 
 Can not modify dirstate outside
 
   $ mkdir outside
   $ touch outside/f1
   $ hg debugwalk -v -I 'relglob:f1'
   * matcher:
   <includematcher includes='(?:|.*/)f1(?:/|$)'>
   f  inside/f1  inside/f1
   $ hg add .
   $ hg add outside/f1
   abort: cannot track 'outside/f1' - it is outside the narrow clone
   [255]
   $ touch outside/f3
   $ hg add outside/f3
   abort: cannot track 'outside/f3' - it is outside the narrow clone
   [255]
 
 But adding a truly excluded file shouldn't count
 
   $ hg add outside/f3 -X outside/f3
 
   $ rm -r outside
 
 Can modify dirstate inside
 
   $ echo modified > inside/f1
   $ touch inside/f3
   $ hg add inside/f3
   $ hg status
   M inside/f1
   A inside/f3
   $ hg revert -qC .
   $ rm inside/f3
 
 Can commit changes inside. Leaves outside unchanged.
 
   $ hg update -q 'desc("initial")'
   $ echo modified2 > inside/f1
   $ hg manifest --debug
   4d6a634d5ba06331a60c29ee0db8412490a54fcd 644   inside/f1
   7fb3bb6356d28d4dc352c5ba52d7350a81b6bd46 644   outside/f1 (flat !)
   d0f2f706468ab0e8bec7af87446835fb1b13511b 755 d outside/ (tree !)
   $ hg commit -m 'modify inside/f1'
   created new head
   $ hg files -r .
   inside/f1
   $ hg manifest --debug
   3f4197b4a11b9016e77ebc47fe566944885fd11b 644   inside/f1
   7fb3bb6356d28d4dc352c5ba52d7350a81b6bd46 644   outside/f1 (flat !)
   d0f2f706468ab0e8bec7af87446835fb1b13511b 755 d outside/ (tree !)
 Some filesystems (notably FAT/exFAT only store timestamps with 2
 seconds of precision, so by sleeping for 3 seconds, we can ensure that
 the timestamps of files stored by dirstate will appear older than the
 dirstate file, and therefore we'll be able to get stable output from
 debugdirstate. If we don't do this, the test can be slightly flaky.
   $ sleep 3
   $ hg status
   $ hg debugdirstate --no-dates
   n 644         10 set                 inside/f1
 
-Can't commit empty files
+Can commit empty files
 
   $ touch inside/c; hg add inside/c; hg commit -qm _; hg verify -q
-   warning: revlog 'data/inside/c.i' not in fncache!
-   4: empty or missing inside/c
-   inside/c@4: manifest refers to unknown revision 000000000000
-  1 warnings encountered!
-  hint: run "hg debugrebuildfncache" to recover from corrupt fncache
-  2 integrity errors encountered!
-  (first damaged changeset appears to be 4)
-  [1]
+  $ hg cat -r . inside/c