Before this patch, immediate value is assigned to dirstate._cwd, if
ui.forcecwd is specified at instantiation of dirstate.
But this doesn't work as expected in some cases.
For example, hgweb set ui.forcecwd after instantiation of repo object.
If an extension touches repo.dirstate in its reposetup(), dirstate is
instantiated without setting ui.forcecwd, and dirstate.getcwd()
returns incorrect result.
In addition to it, hgweb.__init__() can take already instantiated repo
object, too. In this case, repo.dirstate might be already
instantiated, even if all enabled extensions don't so in their own
reposetup().
To avoid such issue, this patch centralizes _cwd handling into _cwd
method.
This issue can be reproduced by running test-hgweb-commands.t with
fsmonitor-run-tests.py.
The general delta heuristic to select a delta do not scale with the number of
branch. The delta base is frequently too far away to be able to reuse a chain
according to the "distance" criteria. This leads to insertion of larger delta (or
even full text) that themselves push the bases for the next delta further away
leading to more large deltas and full texts. This full text and frequent
recomputation throw Mercurial performance in disarray.
For example of a slightly large repository
280 000 files (2 150 000 versions)
430 000 changesets (10 000 topological heads)
Number below compares repository with and without the distance criteria:
manifest size:
with: 21.4 GB
without: 0.3 GB
store size:
with: 28.7 GB
without 7.4 GB
bundle last 15 00 revisions:
with: 800 seconds
971 MB
without: 50 seconds
73 MB
unbundle time (of the last 15K revisions):
with: 1150 seconds (~19 minutes)
without: 35 seconds
Similar issues has been observed in other repositories.
Adding a new option or "feature" on stable is uncommon. However, given that this
issues is making Mercurial practically unusable, I'm exceptionally targeting
this patch for stable.
What is actually needed is a full rework of the delta building and reading
logic. However, that will be a longer process and churn not suitable for stable.
In the meantime, we introduces a quick and dirty mitigation of this in the
'experimental' config space. The new option introduces a way to set the maximum
amount of memory usable to store a diff in memory. This extend the ability for
Mercurial to create chains without removing all safe guard regarding memory
access. The option should be phased out when core has a more proper solution
available.
Setting the limit to '0' remove all limits, setting it to '-1' use the default
limit (textsize x 4).
People often want to know what they are working on *now*. As part of
this, they also commonly want to know how that work is related to other
changesets in the repo so they can perform common actions like rebase,
histedit, and merge.
`hg show work` made headway into this space. However, it is geared
towards a complete repo view as opposed to just the current line of
work. If you have a lot of in-flight work or the repo has many heads,
the output can be overwhelming. The closest thing Mercurial has to
"show me the current thing I'm working on" that doesn't require custom
revsets is `hg qseries`. And this requires MQ, which completely changes
workflows and repository behavior and has horrible performance on large
repos. But as sub-optimal as MQ is, it does some things right, such as
expose a model of the repo that is easy for people to reason about.
This simplicity is why I think a lot of people prefer to use MQ, despite
its shortcomings.
One common development workflow is to author a series of linear
changesets, using bookmarks, branches, anonymous heads, or even topics
(3rd party extension). I'll call this a "stack." You periodically
rewrite history in place (using `hg histedit`) and reparent the stack
against newer changesets (using `hg rebase`). This workflow can be
difficult because there is no obvious way to quickly see the current
"stack" nor its relation to other changesets. Figuring out arguments to
`hg rebase` can be difficult and may require highlighting and pasting
multiple changeset nodes to construct a command.
The goal of this commit is to make stack based workflows simpler
by exposing a view of the current stack and its relationship to
other releant changesets, notably the parent of the base changeset
in the stack and newer heads that the stack could be rebased or merged
into.
Introduced is the `hg show stack` view. Essentially, it finds all
mutable changesets from the working directory revision in both
directions, stopping at a merge or branch point. This limits the
revisions to a DAG linear range.
The stack is rendered as a concise list of changesets. Alongside the
stack is a visualization of the DAG, similar to `hg log -G`.
Newer public heads from the branch point of the stack are rendered
above the stack. The presence of these heads helps people understand
the DAG model and the relationship between the stack and changes made
since the branch point of that stack. If the "rebase" command is
available, a `hg rebase` command is printed for each head so a user
can perform a simple copy and paste to perform a rebase.
This view is alpha quality. There are tons of TODOs documented
inline. But I think it is good enough for a first iteration.
The ignore regular expression has been updated to detect
"inconsistent config." If present, we track which configs have
that set and we suppress the conflicting defaults error for those
options.
I also added named groups to the regexp to aid readability.
A comment was added to profiling.py to make a desired inconsistent
value error go away.
It isn't needed that storecache is derived from repofilecache.
Changes in this patch allow repofilecache and storecache to do in own
__init__() differently from each other.
This is a fix for my 1c4d8237ba73, which used 'bool(dir)' as
checkambig value for revlog.__init__().
I can't remember why I did so in 1c4d8237ba73, but this is obviously
wrong, because only root indexfile is cached via filecache-ed property
of localrepository.
* Scope of "value" is reduced
* index_baserev() is documented
* Error is no longer redundantly set for -2 return values
* Error values are compared <= -2 instead of == -2 to protect
against odd failure scenarios
I've seen revlog._deltachain() appear in a number of performance
profiles. I suspect there are 2 reasons for this:
1. Delta chain resolution performs many index lookups, thus triggering
population of index tuples. Creating possibly tens of thousands of
PyObject will have overhead.
2. Delta chain resolution is a tight loop.
By moving delta chain resolution to C, we can defer instantiation
of full index entry tuples and make the loop faster courtesy of
not running in Python.
We can measure the impact to delta chain resolution via
`hg perflogrevision` using the mozilla-central repo with a recent
manifest having delta chain length of 33726:
$ hg perfrevlogrevision -m 364895
! full
! wall 0.367585 comb 0.370000 user 0.340000 sys 0.030000 (best of 27)
! wall 0.357581 comb 0.360000 user 0.350000 sys 0.010000 (best of 28)
! deltachain
! wall 0.010644 comb 0.010000 user 0.010000 sys 0.000000 (best of 270)
! wall 0.000292 comb 0.000000 user 0.000000 sys 0.000000 (best of 8729)
$ hg perfrevlogrevision --cache -m 364895
! deltachain
! wall 0.003904 comb 0.000000 user 0.000000 sys 0.000000 (best of 712)
! wall 0.000284 comb 0.000000 user 0.000000 sys 0.000000 (best of 9926)
The first test measures savings from both not instantiating index
entries and moving to C. The second test (which doesn't clear the
index caches) essentially isolates the benefits of moving from Python
to C. It still shows a 13.7x speedup (versus 36.4x). And there are
multiple milliseconds of savings within the critical path for resolving
revision data. I think that justifies the existence of C code.
A more striking example of the benefits of this change can be
demonstrated by timing `hg debugdeltachain -m` for the mozilla-central
repo:
$ time hg debugdeltachain -m > /dev/null
before: 1057.4s
after: 503.3s
PyPy2.7 5.8.0: 220.0s
It's worth noting that the C code isn't as optimal as it could be.
We're still instantiating a new PyObject for every revision. A future
optimization would be to reuse the PyObject on the cached index tuple.
We could potentially also get wins by using a memory array of raw
integers. There is also room for a delta chain cache on revlog
instances. Of course, the best optimization is to implement revlog
reading outside of Python so Python doesn't need to be concerned
about the relatively expensive index entries and operations on them.
Let us start registering the existing option. I'm starting with the 'devel'
section because it is full of useful things that are poorly documented. So
registering these will more than other section.
The function lost its last caller in ae209b610844 (patch: replace
functions in fsbackend to use vfs, 2014-06-05) when the callers
started relying on the opener to do the join.
patchbackend() seems to call it on an arbitrary backend, so it seems
to be part of the API. Since all subclasses do something in their
close() methods, I decided to let this one raise an exception rather
than just pass.
Extensions sometimes wants to add other information in the default log output
format (when no templating is used).
Add an empty function named '_exthook' for easing the extension life.
Extensions will be able to wrap this function and collaborate to display
additional information.
Exthook is called after displaying troubles and just before displaying the
files, extra and description.
Add a new test file to test it and not pollute other test files.
One hitch is that sometimes fcd is actually an absentfilectx which does not
expose any mutator functions. In order to still use the context functions,
we look up the underlying workingfilectx to perform the write there.
One alternate way would be to put the write functions on the absentfilectx and
have them pass-through. While this makes the callsites cleaner, we would need
to decide what its getter functions would return after this point, since
returning None for `data` (and True for `isabsent()`) might no longer be
correct after a write. I discussed with Sidd about just having the getters
raise RuntimeErrors after a mutator has been called, but we actually call
isabsent() in merge.py after running the internal merge tools.
Yuya pointed out that using mutable value as the default could be problematic.
To work around this we now support callable object as default value. This
allows for creating new mutable objects on demand when needed.
None of this function has been used in the past 5 years, so I think it is safe
to just kill them. All code accessing rich markers is using 'getmarkers(...)'
instead (or raw markers).
