On 16/10/2016 12:43, Pierre-Yves David wrote:



On 10/08/2016 05:18 PM, Simon Farnsworth wrote:
# HG changeset patch
# User Simon Farnsworth <simon...@fb.com>
# Date 1475939634 25200
#      Sat Oct 08 08:13:54 2016 -0700
# Node ID 8a0c9c0158b3e9574a4571af3dce9978844b825d
# Parent  ee284d7c5faa5d9f17853f51c17883844b985c58
evolve: lock the working copy early in next and prev (issue5244)

Both next and prev depend on a consistent working copy, but were
waiting to
take the lock until they were ready to alter the working copy.

Take the lock before reading the working copy state, and do not
release it
until we're definitely not going to change the working copy.

The fix approach seems good, but the implementation is a bit strangeto
me. See inline.

I can remove the early unlocks, but the goal is to unlock early when we no longer need the locks.


diff --git a/hgext/evolve.py b/hgext/evolve.py
--- a/hgext/evolve.py
+++ b/hgext/evolve.py
@@ -2213,10 +2213,13 @@
     """update to parent revision

     Displays the summary line of the destination for clarity."""
-    if True:
+    wlock = None
+    dryrunopt = opts['dry_run']
+    if not dryrunopt:
+        wlock = repo.wlock()
+    try:
         wkctx = repo[None]
         wparents = wkctx.parents()
-        dryrunopt = opts['dry_run']
         if len(wparents) != 1:
             raise error.Abort('merge in progress')
         if not opts['merge']:
@@ -2246,7 +2249,6 @@
                 ret = hg.update(repo, p.rev())
                 if not ret:
                     tr = lock = None
-                    wlock = repo.wlock()
                     try:
                         lock = repo.lock()
                         tr = repo.transaction('previous')
@@ -2258,13 +2260,22 @@
                         tr.close()
                     finally:
                         lockmod.release(tr, lock, wlock)
+                        wlock = None
+                else:
+                    lockmod.release(wlock)
+                    wlock = None
+
             displayer.show(p)
             return 0
         else:
+            lockmod.release(wlock)
+            wlock = None


Why are we issuing multiple lockmode.release call instead of just using
the final finally clause ?

Each lockmod.release() happens at a point where we no longer need the lock for correctness (and in the old code, didn't have the lock at all).

The aim is to minimise the time we're holding the lock for. I can easily redo this with wlock held for the entire duration of the command if you'd prefer that version?

             for p in parents:
                 displayer.show(p)
             ui.warn(_('multiple parents, explicitly update to one\n'))
             return 1
+    finally:
+        lockmod.release(wlock)

 @command('^next',
          [('B', 'move-bookmark', False,
@@ -2282,10 +2293,13 @@

     Displays the summary line of the destination for clarity.
     """
-    if True:
+    wlock = None
+    dryrunopt = opts['dry_run']
+    if not dryrunopt:
+        wlock = repo.wlock()
+    try:
         wkctx = repo[None]
         wparents = wkctx.parents()
-        dryrunopt = opts['dry_run']
         if len(wparents) != 1:
             raise error.Abort('merge in progress')
         if not opts['merge']:
@@ -2315,7 +2329,6 @@
                 ret = hg.update(repo, c.rev())
                 if not ret:
                     lock = tr = None
-                    wlock = repo.wlock()
                     try:
                         lock = repo.lock()
                         tr = repo.transaction('next')
@@ -2327,15 +2340,23 @@
                         tr.close()
                     finally:
                         lockmod.release(tr, lock, wlock)
+                        wlock = None
+                else:
+                    lockmod.release(wlock)
+                    wlock = None
             displayer.show(c)
             result = 0
         elif children:
+            lockmod.release(wlock)
+            wlock = None
             ui.warn(_("ambigious next changeset:\n"))
             for c in children:
                 displayer.show(c)
             ui.warn(_('explicitly update to one of them\n'))
             result = 1
         else:
+            lockmod.release(wlock)
+            wlock = None
             aspchildren = _aspiringchildren(repo, [repo['.'].rev()])
             if topic:
                 filtered.extend(repo[c] for c in children
@@ -2368,6 +2389,8 @@
                 return result
             return 1
         return result
+    finally:
+        lockmod.release(wlock)

 def _reachablefrombookmark(repo, revs, bookmarks):
     """filter revisions and bookmarks reachable from the given bookmark
diff --git a/tests/fake-editor.sh b/tests/fake-editor.sh
new file mode 100755
--- /dev/null
+++ b/tests/fake-editor.sh
@@ -0,0 +1,3 @@
+#!/bin/sh
+sleep 5
+echo "new desc" >> $1
diff --git a/tests/test-prev-next.t b/tests/test-prev-next.t
--- a/tests/test-prev-next.t
+++ b/tests/test-prev-next.t
@@ -206,3 +206,34 @@
   move:[5] added d
   atop:[6] added b (3)
   working directory is now at 47ea25be8aea
+
+prev and next should lock properly against other commands
+
+  $ hg init repo
+  $ cd repo
+  $ HGEDITOR=${TESTDIR}/fake-editor.sh
+  $ echo hi > foo
+  $ hg ci -Am 'one'
+  adding foo
+  $ echo bye > foo
+  $ hg ci -Am 'two'
+
+  $ hg amend --edit &
+  $ sleep 1
+  $ hg prev
+  waiting for lock on working directory of $TESTTMP/repo held by
process '*' on host '*' (glob)
+  got lock after [4-6] seconds (re)
+  1 files updated, 0 files merged, 0 files removed, 0 files unresolved
+  [0] one
+  $ wait
+
+  $ hg amend --edit &
+  $ sleep 1
+  $ hg next --evolve
+  waiting for lock on working directory of $TESTTMP/repo held by
process '*' on host '*' (glob)
+  1 new unstable changesets
+  got lock after [4-6] seconds (re)
+  move:[2] two
+  atop:[3] one
+  working directory now at a7d885c75614
+  $ wait
_______________________________________________
Mercurial-devel mailing list
Mercurial-devel@mercurial-scm.org
https://urldefense.proofpoint.com/v2/url?u=https-3A__www.mercurial-2Dscm.org_mailman_listinfo_mercurial-2Ddevel&d=DQICaQ&c=5VD0RTtNlTh3ycd41b3MUw&r=mEgSWILcY4c4W3zjApBQLA&m=0ZuyKYuyHM9hbUFUidarDO5LsEegsCBoK_VAKP3aRHQ&s=8758F-YEyhfMSvO5kgMSC6hUxWqBxsGpOIyAeeaePoU&e=



--
Simon Farnsworth
_______________________________________________
Mercurial-devel mailing list
Mercurial-devel@mercurial-scm.org
https://www.mercurial-scm.org/mailman/listinfo/mercurial-devel

Reply via email to