Yuya Nishihara gave this suggestion on the mailing list after the
previous patch was queued, and honestly this seems much simpler and
looks more efficient.
This renames the spanset class to _spanset, and moves its __init__ to new
spanset() function. spanset() is now a factory function.
This allows us to construct a spanset without keeping a repo instance.
Inspired by the dirstate fix in 39954a8760cd, this should fix any race
conditions with the fsmonitor state changing from underneath.
Since we now grab the wlock for any non-invalidate writes, the only situation
this appears to happen in is with a concurrent invalidation. Test that.
Some extensions like fsmonitor need to run code after dirstate.status is
called, but while the wlock is held. The extensions could grab the wlock again,
but that has its own peculiar race issues. For example, fsmonitor would not
like its state to be written out if the dirstate has changed underneath (see
issue5581 for what can go wrong in that sort of case).
To protect against these sorts of issues, allow extensions to declare that they
would like to run some code to run at fixup time.
fsmonitor will switch to using this in the next patch in the series.
Context manager has a mechanism to control extension propagation. It is not
used by profiling right now, but making the code correct will help prevent bug
in the future.
There are no way for this to happen today, but better be safe than sorry, no
one know how the code will evolve. We now make sure the file pointer is closed
even is profiler is None.
This replaces 'if y in subset' with '& subset'. first(null) and last(wdir())
are fixed thanks to fullreposet.__and__.
This also revealed that first() and last() don't follow the order of the
input set. 'ls & subset' is valid only if the ordering requirement is 'define'
or 'any'.
No performance regression observed:
revset #0: limit(0:9999, 100, 9000)
0) 0.001164
1) 0.001135
revset #2: 9000 & limit(0:9999, 100, 9000)
0) 0.001224
1) 0.001181
revset #3: last(0:9999, 100)
0) 0.000237
1) 0.000199
last() is implemented using a reversed iterator, so the result should be
reversed again.
I've marked this as BC since it's quite old bug seen in 3.0. The first bad
revision is 1ef0875a62f8 "revset: changed last implementation to use lazy
classes."
Instead of walking all ancestors to compute the full set we now check membership
lazily. This massively speed.
On a million-ish revision repository, this remove 14 seconds from the push logic,
making the checkheads function disappear from profile.
The obsolete post-processing needs to know the extend of the pushed set. The
way it is implemented is... suboptimal. It build a full set of all nodes in the
pushset and it does so using changectx. We have much better API for this now.
The simplest is to use the existing lazy ancestors computation. That logic uses
revnum and not node (for good reason) so we start with updating the
post-processing code to handle a "allfuturecommon" set containing revision
numbers.
We are about to call 'torev' on node that might be locally missing. In this
case, 'nodemap.revs' will return None (something valid in our usecase) while
'changelog.rev' would raise an exception.
We make this change in a distinct changeset to show it does not impact the
tests.
We are about to make "allfuturecommon" a set of revs instead of a set of nodes.
The function updated in this patch do not needs to know about these details so
we just pass it a 'ispushed(node)' function. This will simplify the next
changeset.
It seems sufficiently simple to use "profile(enabled=X)" to not justify having
a dedicated context manager just to read the config.
(I do not have a too strong opinion about this).
We now process the "--profile" a second time after alias has been processed and
the command argument fully parsed. If appropriate we enable profiling at that
time.
In these situation, the --profile will cover less than if the full --profile
flag was passed on the command line. This is better than the previous behavior
(flag ignored) and still fullfil multiple valid usecases.
This is a step toward allowing context where the profiling in enabled
withing the context range.
This also open the way to kill the dedicated "maybeprofile" context manager
and keep only one of 'profile' and 'maybeprofile'.
The start method is doing all profiler setup and activation. It is currently
unconditionally called by '__init__' but this will be made more flexible in
later changesets.
So far we have been able to use a simple decorator for this. However using the
current context manager makes the scope of the profiling in dispatch
constrainted and the time frame to decide to enable profiling quite limited
(using "maybeprofile")
This is the first step toward the ability to enable the profiling from within
the profiling scope. eg::
with maybeprofiling(ui) as profiler:
...
bar.foo():
...
if options['profile']:
profiler.start()
...
fooz()
...
My target usecase is adding support for "--profile" to alias definitions with
effect. These are to be used with "profiling.output=blackbox" to gather data
about operation that get slow from time to time (eg: pull being minutes instead
of seconds from time to time).
Of course, in such case, the scope of the profiling would be smaller since
profiler would be started after running extensions 'reposetup' (and other
potentially costly logic), but these are not relevant for my target usecase
(multiple second commits, multiple tens of seconds pull).
Currently adding '--profile' to a command through alias requires to re-spin a
Mercurial binary (using "!$HG" in alias), which as a significant performance
impact, especially in context where startup performance is being worked on...
An alternative approach would be to stop using the context manager in dispatch
and move back to a try/finally setup.
This can happen if another process (even another hg process!) comes along and
removes the file at that time.
This partly resolves issue5584, but not completely -- a bogus dirstate update
can still happen. However, the full fix is too involved for stable.
The 'srvheads' list contains all server heads including the common ones. We
adjust 'ui.log' message to provide more useful information about server heads
locally unknown. The performance impact of turning the list to set is
negligible (about 1e-4s) compared to the rest of the discovery cost, so I'm
taking the easy path.
gitweb was missing the hint hover box. So that was added.
Also, the positioning of the form was absolute and it didn't
vertically align on all pages. The element has been moved inline
with the navigation links (which now are contained in a div) and
flexbox is used to obtain sane alignment of the navigation links
and search form. For those new to flexbox,
"justify-content: space-between" basically says to maximize space
elements. You can use it to easily get left and right justified
containers without having to worry about width, floating, etc.
"align-items: center" centers all items in a cross-axis. I've
literally wasted hours trying to figure out both these problems
before flexbox. Flexbox is amazing.
Flexbox has been supported by Chrome and Firefox for a few years.
But it is only supported by IE 11. I'm willing to wager that
people using this either won't be using IE or will be using IE 11.
So I'm willing to be a bit aggressive in adopting flexbox because
it makes CSS alignment so much easier.
Paper has it on all pages. Not sure why gitweb doesn't. I think it
should be everywhere because it is a useful feature.
Also, we weren't consistently adding the HTML in the same place. This
was OK since the element is absolutely positioned. But this bothered
me a bit, so I went ahead and fixed it.
