Preparing the data for some hooks can be expensive. Add a function to check if
a hook exists so we can skip useless preparation if no hook is configured.
When bootstrapping a deb/rpm build, packagelib.sh starts performing a local
build for the sole purpose of parsing the output of "hg version".
Then it "hg archive"s the source code, and builds everything again.
For that initial step, we are perfectly good in using a pure python mercurial,
without compiling the c modules (base85, bdiff, zstdlib, ...).
On my personal system, this cuts down 22 seconds for a package build (the
bootstrapping build goes from ~30 to ~8 seconds).
Turn dirnode's methods into generators which can be used to update "tersedict"
in caller. So instead of passing the "tersedict" to be mutated here and there,
it's now clearer where it is updated as it's purely a local variable to
tersedir() function.
While I was here, I renamed _processtersestatus to tersewalk and
_addfilestotersed to iterfilepaths.
Differential Revision: https://phab.mercurial-scm.org/D1043
The previous terse status implementation was hacking around os.listdir() and was
flaky. There have been a lot of instances of mercurial buildbots failing
and google's internal builds failing because of the
hacky implementation of terse status. Even though I wrote the last
implementation but it was hard for me to find the reason for the flake.
The new implementation can be slower than the old one but is clean and easy to
understand.
In this we create a node object for each directory and create a tree
like structure starting from the root of the working copy. While building the
tree like structure we store some information on the nodes which will be helpful
for deciding later whether we can terse the dir or not.
Once the whole tree is build we traverse and built the list of files for each
status with required tersing.
There is no behaviour change as the old test, test-status-terse.t passes with
the new implementation.
Differential Revision: https://phab.mercurial-scm.org/D985
In future patches, we may halt the merge process based on configuration or
user requests by raising exceptions. We need to ensure that the mergestate
is unconditionally committed even when such an exception is raised.
Depends on D930.
Differential Revision: https://phab.mercurial-scm.org/D931
Currently, failed file merges will nevertheless force the user to continue
attempting to merge all additional unresolved files. Future patches will allow
the user to halt the merge process instead. This patch first introduces a test
demonstrating the current bejhavior so the upcoming changes are more obvious.
Differential Revision: https://phab.mercurial-scm.org/D930
Now that dirstatemap is the source of truth for the list of directories, let's
move _dirfoldmap on to it.
This pattern of moving cached variables onto the dirstate map makes it easier to
invalidate them, as seen by how the cache invalidation functions are slowly
shrinking to just be recreating the dirstatemap instance.
Differential Revision: https://phab.mercurial-scm.org/D983
Now that dirs is source of truthed on the dirstatemap, let's get rid of the
_dirs propertycache on the dirstate.
Differential Revision: https://phab.mercurial-scm.org/D982
Now that the filefoldmap is source of truthed on the dirstatemap, let's get rid
of the property cache on the dirstate.
Differential Revision: https://phab.mercurial-scm.org/D981
Moving the identity function to the dirstatemap class will allow alternative
dirstate implementations to replace the implementation.
Differential Revision: https://phab.mercurial-scm.org/D980
As part of separating dirstate business logic from storage, let's move the
nonnormal and otherparent storage to the dirstatemap class. This will allow
alternative dirstate storage to persist these sets instead of recomputing them.
Differential Revision: https://phab.mercurial-scm.org/D979
As part of separating the dirstate business logic from the dirstate storage
logic, let's move the serialization code down into dirstatemap.
Differential Revision: https://phab.mercurial-scm.org/D978
As part of separating the dirstate business logic from the storage, let's move
the read code into the new dirstatemap class.
Differential Revision: https://phab.mercurial-scm.org/D977
Some section can contains arbitrary keys (eg: color, alias, extensions). We
add a way to register some generic config items for them. This is necessary to
get all the config registered. We use a regular expression because some sub-
attributes (eg: hooks.xxx.priority) can define default value on their own.
Upon pull or unbundle, we display a message with the range of new revisions
fetched. This revision range could readily be used after a pull to look out
what's new with 'hg log'. The algorithm takes care of filtering "obsolete"
revisions that might be present in transaction's "changes" but should not be
displayed to the end user.
This is a simpler workaround alternative to D958, "i18n: clean msgcache when
encoding changes." The cache won't be bloated unless you run tons of commands
with different --encoding options on command server, or serve many repositories
of different web.encoding options on hgweb.
The test was originally written by Jun Wu.
Differential Revision: https://phab.mercurial-scm.org/D1053
This was unintentionally changed in b8023e389b64. Since ifcontains() takes
needle of any types, it shouldn't abort depending on the given container type.
This rule is no longer useful because chg daemon may be killed and respawned
per config/environment hash. We can't reliably run a daemon in foreground.
This changeset introduces a new documented test case for a distributed
obsolescence scenario. The scenario involves a simple case where some
obsolescence markers are retrieved before the changeset they affect. See the
test case documentation for details.
We also test variants where the changesets are added from a bundle.
This test file introduces documented test case for obsolescence markers usage
that are important to distributed workflow cases.
In the distributed case, new changesets, markers, and phases can be added in
orders that would not happen during the local only usage. Documenting these
scenarios and test them is important as we make progress with various
obsolescence aspects.
In repoview.py, computeunserved() and computemutable() functions had the same
documentation. The documentation of computemutable() is wrong. I was unable to
write documentation for the function but it's better to not having the
documentation than having it wrong.
Differential Revision: https://phab.mercurial-scm.org/D1016
This is done by:
sed -i "s/pycompat\.sysplatform == 'darwin'/pycompat.isdarwin/" **/*.py
Plus a manual change to `sslutil.py` which involves indentation change that
cannot be done by `sed`.
Differential Revision: https://phab.mercurial-scm.org/D1035
As suggested by Ryan in D1019, it's cleaner if we use defined constants
instead of `osname == 'nt'` everywhere.
Differential Revision: https://phab.mercurial-scm.org/D1033
With hgdemandimport disabled (chg's case), `import uuid` has an immediate
side effect calling `ctypes.util.find_library` trying to locate the
`libuuid` library. This happens at `import` time before `dispatch.run()`.
The call trace is like:
File "hg/hg", line 54, in <module>
from mercurial import (
File "hg/mercurial/dispatch.py", line 24, in <module>
from . import (
File "hg/mercurial/commands.py", line 23, in <module>
from . import (
File "hg/mercurial/help.py", line 33, in <module>
from .hgweb import (
File "hg/mercurial/hgweb/__init__.py", line 20, in <module>
from . import (
File "hg/mercurial/hgweb/hgweb_mod.py", line 14, in <module>
from .common import (
File "hg/mercurial/hgweb/common.py", line 15, in <module>
import uuid
File "/usr/lib64/python2.7/uuid.py", line 404, in <module>
lib = ctypes.CDLL(ctypes.util.find_library(libname))
The problem is, `ctypes.util.find_library` will execute
`sh -c '/sbin/ldconfig -p 2>/dev/null'` on Python <= 2.7.12. The output of
`sh` may pollute the terminal:
shell-init: error retrieving current directory: getcwd: cannot access
parent directories: No such file or directory
This patch moves `import uuid` so its side-effect can only happen after the
cwd check in `dispatch._getlocal`. Therefore the terminal won't be
polluted by importing `uuid`.
Differential Revision: https://phab.mercurial-scm.org/D1024
