These classes are pretty large and independent from revset computation.
2961 mercurial/revset.py
973 mercurial/smartset.py
3934 total
revset.prettyformatset() is renamed to smartset.prettyformat(). Smartset
classes are aliased since they are quite common in revset.py.
It was probably unintentional for regex, as the meaning of some sequences like
\S and \s is actually inverted by changing the case. For backward compatibility
however, the matching is forced to case insensitive.
Let's resurrect the docstring since our help module can detect the EXPERIMENTAL
tag and display it only if -v is specified.
This patch updates the test added by bbdfa2d5aaa2 since wdir() is now
documented.
Unlike p1 = null, p2 = null denotes the revision has only one parent, which
shouldn't be considered a child of the null revision. This was spotted while
fixing the issue4682 and rediscovered as issue5439.
Because smartset.reverse() may modify the underlying subset, it should be
called only if the set can define the ordering.
In the following example, 'a' and 'c' is the same object, so 'b.reverse()'
would reverse 'a' unexpectedly.
# '0:2 & reverse(all())'
<filteredset
<spanset- 0:2>, # a
<filteredset # b
<spanset- 0:2>, # c
<spanset+ 0:9>>>
present() is special in that it returns the argument set with no
modification, so the ordering requirement should be forwarded.
We could make present() fix the order like orset(), but that would be silly
because we know the extra filtering cost is unnecessary.
This fixes the order of 'x & (y + z)' where 'y' and 'z' are trivial, and the
other uses of _list()-family functions. The original functions are renamed to
'_ordered(|int|hex)list' to say clearly that they do not follow the subset
ordering.
This fixes the order of 'x & (y + z)' where 'y' and 'z' are not trivial.
The follow-order 'or' operation is slower than the ordered operation if
an input set is large:
#0#1#2#3
0) 0.002968 0.002980 0.002982 0.073042
1) 0.004513 0.004485 0.012029 0.075261
#0: 0:4000 & (0:1099 + 1000:2099 + 2000:3099)
#1: 4000:0 & (0:1099 + 1000:2099 + 2000:3099)
#2: 10000:0 & (0:1099 + 1000:2099 + 2000:3099)
#3: file("path:hg") & (0:1099 + 1000:2099 + 2000:3099)
I've tried another implementation, but which appeared to be slower than
this version.
ss = [getset(repo, fullreposet(repo), x) for x in xs]
return subset.filter(lambda r: any(r in s for s in ss), cache=False)
This makes the number of 'or' arguments deterministic so we can attach
additional ordering flag to all operator nodes. See the next patch.
We rewrite the tree immediately after chained 'or' operations are flattened
by simplifyinfixops(), so we don't need to care if arguments are stored in
x[1] or x[1:].
"-p <stage>" is useful for investigating parsing stages. With -p option, a
transformed tree is printed no matter if it is changed or not, which allows
us to know valid stage names by "-p all".
This will allow us to evaluate unoptimized tree and compare the result with
optimized one.
The private _analyze() function isn't renamed since I'll add more parameters
to it.
Before, a keyvalue node was processed by the last catch-all condition of
_optimize(). Therefore, topo.firstbranch=expr would bypass tree rewriting
and would crash if an expr wasn't trivial.
Before the error was caught at func() as an unknown identifier, and the
optimizer failed to detect the syntax error. This patch introduces getsymbol()
helper to ensure that a string is not allowed as a function name.
The ordering of 'x & head()' was broken in 329d82866742 (revset:
improve head revset performance, 2014-03-13). Presumably due to other
optimizations since then, undoing that change to fix the order does
not slow down the simple case of "hg log -r 'head()'" mentioned in
that commit. I see a small slowdown from ~0.16s to about ~0.19s with
'not 0 & head()', but I'd say it's worth it for the correct output.
This adds mostly broken tests that will be fixed by subsequent patches. We
generally don't do that, but this patch series would be hard to review
without a set of broken tests.
Note that some tests pass thanks to the reordering problem in optimize().
For instance, '2:0 & _intlist(0 1 2)' doesn't fail because it is rewritten
as '_intlist(0 1 2) & 2:0'.
Make it noop as before ddf6bfe09ab2. We could change it to an error, but
allowing empty key makes some sense for scripting that builds a key string
programmatically.
Sort revisions in reverse revision order but grouped by topographical branches.
Visualised as a graph, instead of:
o 4
|
| o 3
| |
| o 2
| |
o | 1
|/
o 0
revisions on a 'main' branch are emitted before 'side' branches:
o 4
|
o 1
|
| o 3
| |
| o 2
|/
o 0
where what constitutes a 'main' branch is configurable, so the sort could also
result in:
o 3
|
o 2
|
| o 4
| |
| o 1
|/
o 0
This sort was already available as an experimental option in the graphmod
module, from which it is now removed.
This sort is best used with hg log -G:
$ hg log -G "sort(all(), topo)"
9981d464ac53 fixed matching() to preserve the order of the input set, but
the test was incorrect. Given "A and B", "A" should be the input set to "B".
But thanks to our optimizer, the test expression was rewritten as
"(2 or 3 or 1) and matching(1 or 2 or 3)", therefore it was working well.
Since I'm going to fix the overall ordering issue, the test needs to be
adjusted to do the right thing.
Unlike range, dagrange has no inverted range (such as '10:0'). So there should
be no practical reason to keep dagrange as a function that forces its own
ordering.
No performance regression is spotted in contrib/base-revsets.txt.
Our invert() function was too clever to not take length into account. I could
fix the problem by appending '\xff' as a terminator (opposite to '\0'), but
it turned out to be slower than simple multi-pass sorting.
New implementation is pretty straightforward, which just calls sort() from the
last key. We can do that since Python sort() is guaranteed to be stable. It
doesn't sound nice to call sort() multiple times, but actually it is faster.
That's probably because we have fewer Python codes in hot loop, and can avoid
heavy string and list manipulation.
revset #0: sort(0:10000, 'branch')
0) 0.412753
1) 0.393254
revset #1: sort(0:10000, '-branch')
0) 0.455377
1) 0.389191 85%
revset #2: sort(0:10000, 'date')
0) 0.408082
1) 0.376332 92%
revset #3: sort(0:10000, '-date')
0) 0.406910
1) 0.380498 93%
revset #4: sort(0:10000, 'desc branch user date rev')
0) 0.542996
1) 0.486397 89%
revset #5: sort(0:10000, '-desc -branch -user -date -rev')
0) 0.965032
1) 0.518426 53%
Cpython and pypy have different way to build and order set, so the result of
list(myset) is different. We work around this by using the sorted version of the
data when displaying a list.
This get pypy closer to pass test-revset.t.
Since _parsealiasdefn() rejects unknown alias arguments, _checkaliasarg() is
unnecessary. New test is added to make sure unknown '$n' symbols are rejected.
In short, this patch moves the hack from tokenizedefn() to _relabelaliasargs(),
which is called after parsing. This change aims to eliminate tight dependency
on the revset tokenizer.
Before this patch, we had to rewrite an alias argument to a pseudo function:
"$1" -> "_aliasarg('$1')"
('symbol', '$1') -> ('function', ('symbol', '_aliasarg'), ('string', '$1'))
This was because the tokenizer must generate tokens that are syntactically
valid. By moving the process to the parsing phase, we can assign a unique tag
to an alias argument.
('symbol', '$1') -> ('_aliasarg', '$1')
Since new _aliasarg node never be generated from a user input, we no longer
have to verify a user input at findaliases(). The test for _aliasarg("$1") is
removed as it is syntactically valid and should pass the parsing phase.
I'm going to refactor the alias processing functions. We need '_aliasarg' tag
to limit the scope of the alias expansion, but it wasn't covered by the test.
This patch adds the test that should fail if '_aliasarg' were 'symbol'.
This is the first half of the second part of the "template alias" series. The
whole series will consist of the following parts:
1. make parsed template tree to be compatible with parser functions
(98f4895a3a97 and d7aaf6f4715f)
2. refactor alias processing to be less dependent on revset module
(1/2 in this series)
3. extract reusable component to parser module
4. clean up it
5. extend it to support template syntax
6. add debugging/testing functions of template aliases
7. add alias expansion routine to templater
A filteredset is heavily used, but it cannot provide a printable information
how given set is filtered because a condition is an arbitrary callable object.
This patch adds an optional "condrepr" object that is used only by repr(). To
minimize the maintaining/runtime overhead of "condrepr", its type is overloaded
as follows:
type example
-------- ---------------------------------
tuple ('<not %r>', other)
str '<branch closed>'
callable lambda: '<branch %r>' % sorted(b)
object other
This patch consists of changes below (these can't be applied
separately).
- replace revset.extpredicate by registrar.revsetpredicate in
extensions
- remove setup() on an instance named as revsetpredicate in
uisetup()/extsetup() of each extensions
registrar.revsetpredicate doesn't have setup() API.
- put new entry for revsetpredicate into extraloaders in dispatch
This causes implicit loading predicate functions at loading
extension.
This loading mechanism requires that an extension has an instance
named as revsetpredicate, and this is reason why
largefiles/__init__.py is also changed in this patch.
Before this patch, test-revset.t tests that all decorated revset
predicates are loaded by explicit setup() at once ("all or nothing").
Now, test-revset.t tests that any revset predicate isn't loaded at
failure of loading extension, because loading itself is executed by
dispatch and it can't be controlled on extension side.
Previously, revsets like 'X - Y' were translated to be 'X and not Y'. This can
be expensive, since if Y is a single commit then 'not Y' becomes a huge set and
sometimes the query optimizer doesn't account for it well.
This patch changes revsets to use the built in smartset minus operator, which is
often smarter than 'X and not Y'.
On a large repo this saves 2.2 seconds on rebase and histedit because "X:: - X"
becomes almost instant.
Relevant performance numbers from revsetbenchmark.py
revset #13: roots((tip~100::) - (tip~100::tip))
plain min max first last reverse rev..rst rev..ast sort sor..rst sor..ast
0) 0.001080 0.001107 0.001102 0.001118 0.001121 0.001114 0.001141 0.001123 0.001099 0.001123 0.001137
1) 0.000708 65% 0.000738 66% 0.000735 66% 0.000739 66% 0.000784 69% 0.000780 70% 0.000807 70% 0.000756 67% 0.000727 66% 0.000759 67% 0.000808 71%
revset #14: roots((0::) - (0::tip))
plain min max first last reverse rev..rst rev..ast sort sor..rst sor..ast
0) 0.131304 0.079168 0.133129 0.076560 0.048179 0.133349 0.049153 0.077097 0.129689 0.076212 0.048543
1) 0.065066 49% 0.036941 46% 0.066063 49% 0.034755 45% 0.048558 0.071091 53% 0.047679 0.034984 45% 0.064572 49% 0.035680 46% 0.048508
revset #22: (not public() - obsolete())
plain min max first last reverse rev..rst rev..ast sort sor..rst sor..ast
0) 0.000139 0.000133 0.000133 0.000138 0.000134 0.000155 0.000157 0.000152 0.000157 0.000156 0.000153
1) 0.000108 77% 0.000129 0.000129 0.000134 0.000132 0.000127 81% 0.000151 0.000147 0.000127 80% 0.000152 0.000149
revset #25: (20000::) - (20000)
plain min max first last reverse rev..rst rev..ast sort sor..rst sor..ast
0) 0.050560 0.045513 0.022593 0.043588 0.021909 0.045517 0.021822 0.044660 0.049740 0.044227 0.021819
1) 0.018614 36% 0.000171 0% 0.019659 87% 0.000168 0% 0.015543 70% 0.021069 46% 0.015623 71% 0.000180 0% 0.018658 37% 0.000186 0% 0.015750 72%
Internal _matchfiles() function can take bunch of arguments, which would
lead to a maximum recursion depth error. This patch avoids the excessive
stack use by flattening 'list' nodes beforehand.
Since getlist() no longer takes a nested 'list' nodes, _parsealiasdecl()
also needs to flatten argument list, "aliasname($1, $2, ...)".
Before this patch, test-revset.t fails on Solaris using ksh as default
"sh", because nested quoting below isn't acceptable for it.
+-----------------------------+ inner quoting
"`python -c "print '|'.join(['0:1'] * 500)"`"
+-------------------------------------------+ outer quoting
This patch does below for portability.
- omit outer quoting
This should be safe, because generated string contains no white
space character.
- use '+' instead of '|' (for safety)
'|' has special meaning for many shell, but '+' isn't (at least,
for ordinary ones).
Previous patch introduced 'revset.predicate' decorator to register
revset predicate function easily.
But it shouldn't be used in extension directly, because it registers
specified function immediately. Registration itself can't be restored,
even if extension loading fails after that.
Therefore, registration should be delayed until 'uisetup()' or so.
This patch uses 'extpredicate' decorator derived from 'delayregistrar'
to register predicate in extension easily.
This patch also tests whether 'registrar.delayregistrar' avoids
function registration if 'setup()' isn't invoked on it, because
'extpredicate' is the first user of it.
It's common for GUI or web frontend to fetch chunk of revisions per batch
size. Previously it was possible only if revisions were sorted by revision
number.
$ hg log -r 'limit({revspec} & :{last_known}, 101)'
So this patch introduces a general way to retrieve chunk of revisions after
skipping offset revisions.
$ hg log -r 'limit({revspec}, 100, {last_count})'
This is a dumb implementation. We can optimize it for baseset and spanset
later.
smartset sorting is lazy (so faster in some case) and better (informs that the
set is sorted allowing some optimisation). So we rely on it directly.
Some test output are updated because we now have more information (ordering).
The odd-range hack was introduced by 9afcfbca1710 for backward compatibility,
but it was too widely applied. I've checked cmdutil.revpair() at 1.6, and
found that ".:", ":0" and ":" are also handled as pairs. So let's enable the
hack only for "x:y", "x:", "y:" and ":".
test-revset.t is updated because "tip^::tip^ or tip^" shouldn't be taken as
an odd range. This patch adds "tip^:tip^" instead.
This patch is written for the default branch because parse() of the stable
branch lacks compatibility hack for "foo+bar" tag. If we want to mitigate the
issue in stable, we can add something like "and '::' in revs[0]".
An empty group expression "()" generates None in AST, so it should be tested
before destructuring a tuple.
"A | ()" is still evaluated to an error because I'm not sure whether "()"
represents an empty set or an empty expression (= a unit value). They are
identical in "or" operation, but they should be evaluated differently in
"and" operation.
expression empty set unit value
---------- --------- ----------
() {} A
A & () {} A
A | () A A
An empty group expression "()" generates None in AST, so the optimizer have
to test it before destructuring a tuple. The error message, "missing argument",
is somewhat obscure, but it should be better than crash.
The next patch will move the exceptional parsing of old-style ranges
to revset.tokenize(). This patch will allow us to see the result tree.
Note that the parsing result of '-a-b-c-' is incorrect at this changeset.
It will be fixed soon.
This is the simplest way to handle wdir() revision in revset. None didn't
work well because revset heavily depends on integer operations such as min(),
max(), sorted(), x:y, etc.
One downside is that we cannot do "wctx.rev() in set" because wctx.rev() is
still None. We could wrap the result set by wdirproxyset that translates None
to wdirrev, but it seems overengineered at this point.
result = getset(repo, subset, tree)
if 'wdir' in funcsused(tree):
result = wdirproxyset(result)
Test cases need the '(all() + wdir()) &' hack because we have yet to fix the
bootstrapping issue of null and wdir.
It will be used as an keyword argument.
Note that our "=" operator is left-associative. In general, the assignment
operator is right-associative, but we don't care because it isn't allowed to
chain "=" operations.
I noticed when I mistyped 'matching', that it suggested '_matchfiles' as well.
Rather than simply exclude names that start with '_', this excludes anything
without a docstring. That way, if it isn't in the help text, it isn't
suggested, such as 'wdir()'.
This reduces the stack depth from O(n) to O(log(n)). Therefore, repeated
-rREV options will never exceed the Python stack limit.
Currently it depends on private revset._combinesets() function. But at some
point, we'll be able to drop the old-style parser, and revrange() can be
completely rewritten without using _combinesets():
trees = [parse(s) for s in revs]
optimize(('or',) + trees) # combine trees and optimize at once
...
Blockers that prevent eliminating old-style parser:
- nullary ":" operator
- revrange(repo, [intrev, ...]), can be mapped to 'rev(%d)' ?
- revrange(repo, [binnode, ...]), should be banned ?
As seen in issue4565 and issue4624, GUI wrappers and automated scripts are
likely to generate a long query that just has numeric revisions joined by 'or'.
One reason why is that they allows users to choose arbitrary revisions from
a list. Because this use case isn't handled well by smartset, let's optimize
it to a plain old list.
Benchmarks:
1) reduce nesting of chained 'or' operations
2) optimize to a list (this patch)
revset #0: 0 + 1 + 2 + ... + 1000
1) wall 0.483341 comb 0.480000 user 0.480000 sys 0.000000 (best of 20)
2) wall 0.025393 comb 0.020000 user 0.020000 sys 0.000000 (best of 107)
revset #1: sort(0 + 1 + 2 + ... + 1000)
1) wall 0.035240 comb 0.040000 user 0.040000 sys 0.000000 (best of 100)
2) wall 0.026432 comb 0.030000 user 0.030000 sys 0.000000 (best of 102)
revset #2: first(0 + 1 + 2 + ... + 1000)
1) wall 0.028949 comb 0.030000 user 0.030000 sys 0.000000 (best of 100)
2) wall 0.025503 comb 0.030000 user 0.030000 sys 0.000000 (best of 106)
This reduces the stack depth of chained 'or' operations:
- from O(n) to O(1) at the parsing, alias expansion and optimization phases
- from O(n) to O(log(n)) at the evaluation phase
simplifyinfixops() must be applied immediately after the parsing phase.
Otherwise, alias expansion would crash by "maximum recursion depth exceeded"
error.
Test cases use 'x:y|y:z' instead of 'x|y' because I'm planning to optimize
'x|y' in a different way.
Benchmarks:
0) a2acce8dcd95
1) this patch
revset #0: 0 + 1 + 2 + ... + 200
0) wall 0.026347 comb 0.030000 user 0.030000 sys 0.000000 (best of 101)
1) wall 0.023858 comb 0.030000 user 0.030000 sys 0.000000 (best of 112)
revset #1: 0 + 1 + 2 + ... + 1000
0) maximum recursion depth exceeded
1) wall 0.483341 comb 0.480000 user 0.480000 sys 0.000000 (best of 20)
revset #2: sort(0 + 1 + 2 + ... + 200)
0) wall 0.013404 comb 0.010000 user 0.010000 sys 0.000000 (best of 196)
1) wall 0.006814 comb 0.010000 user 0.010000 sys 0.000000 (best of 375)
revset #3: sort(0 + 1 + 2 + ... + 1000)
0) maximum recursion depth exceeded
1) wall 0.035240 comb 0.040000 user 0.040000 sys 0.000000 (best of 100)
This warning exists to prevent git users from prematurely polluting
their namespace when trying out Mercurial. But for repos that already
have multiple branches, understanding what branches are is not
optional so we should just shut up.
This fixes the crash caused by "branch(null)" revset. No cache should be
necessary for nullrev because changelog.branchinfo(nullrev) does not involve
IO operation.
Note that the problem of "branch(wdir())" isn't addressed by this patch.
"wdir()" will raise TypeError in many places because of None. This is the
reason why "wdir()" is still experimental.
This patch partially backs out a9a86cbbc5b2 and adds an alternative workaround
to functions that evaluate "null" and "wdir()". Because the new workaround is
incomplete, "first(null)" and "min(null)" don't work as expected. But they were
not usable until 3.4 and "null" isn't commonly used, we can postpone a complete
fix for 3.5.
The issue4682 was caused because "branch(default)" is evaluated to
"<filteredset <fullreposet>>", keeping fullreposet magic. The next patch will
fix crash on "branch(null)", but without this patch, it would make
"null in <branch(default)>" be True, which means "children(branch(default))"
would return all revisions but merge (p2 != null).
I believe the right fix is to stop propagating fullreposet magic on filter(),
but it wouldn't fit to stable release. Also, we should discuss how to handle
"null" and "wdir()" in revset before.
The patch solves two issues:
1. id(unknown_full_hash) aborts, but id(unknown_short_hash) doesn't
2. id(40byte_tag_or_bookmark) returns tagged/bookmarked revision,
but id(non-40byte_tag_or_bookmark) doesn't
After the patch:
1. id(unknown_full_hash) doesn't abort
2. id(40byte_tag_or_bookmark) returns empty set
With "dash" (as "/bin/sh" on Debian GNU/Linux), command execution in
"ENV=val foo bar" style doesn't work as expect in test script files,
if "foo" is user-defined function: it works fine, if "foo" is existing
commands like "hg".
5acecb004820 introduced tests for HGPLAIN and HGPLAINEXCEPT into
test-revset.t, and all of them are in such style.
This patch doesn't:
- add explicit unsetting for HGPLAIN and HGPLAINEXCEPT
they are already introduced by 5acecb004820
- write assignment and exporting in one line
"ENV=val; export ENV" for two or more environment variables in one
line causes failure of test-check-code-hg.t: it is recognized as
"don't export and assign at once" unfortunately.
ui.plain() is supposed to disable config options that change the UI to the
detriment of scripts. As the test demonstrates, revset aliases can actually
override builtin ones, just like command aliases. Therefore I believe this is a
bugfix and appropriate for stable.
I came across a case where an internal command was using a revset that I didn't
immediately pass in and it was difficult to debug what was going wrong with the
revset. This prints out the revset and informs the user that the error is with
a rebset so it should be more obvious what and where the error is.
If self is a smartset and other is a fullreposet, nothing should be necessary.
A small win for trivial query in mozilla-central repo:
revset #0: (0:100000)
0) wall 0.017211 comb 0.020000 user 0.020000 sys 0.000000 (best of 163)
1) wall 0.001324 comb 0.000000 user 0.000000 sys 0.000000 (best of 2160)
The main purpose of wdir() is to annotate working-directory files.
Currently many commands and revsets cannot handle workingctx and may raise
exception. For example, -r ":wdir()" results in TypeError. This problem will
be addressed by future patches.
We could add "wdir" symbol instead, but it would conflict with the existing
tag, bookmark or branch. So I decided not to.
List of commands that will potentially support workingctx revision:
command default remarks
-------- ------- -----------------------------------------------------
annotate p1 useful
archive p1 might be useful
cat p1 might be useful on Windows (no cat)
diff p1:wdir (default)
export p1 might be useful if wctx can have draft commit message
files wdir (default)
grep tip:0 might be useful
identify wdir (default)
locate wdir (default)
log tip:0 might be useful with -p or -G option
parents wdir (default)
status wdir (default)
This patch includes minimal test of "hg status" that should be able to handle
the workingctx revision.
Before this patch, when I have a brain fart and type `hg log -r
'add(foo)'`, hg exits and just says add isn't a function, leading me
to the help page for revset to figure out how to spell the
function. With this patch, it suggests 'adds' as a function I might
have meant.
As per fullreposet.__and__, it can omit the range check of rev. Therefore,
"null" revision is accepted automagically.
It seems this can fix many query results involving null symbol. Originally,
the simplest "(null)" query did fail if there were hidden revisions. Tests
are randomly chosen.
fullreposet mimics the behavior of localrepo, where "null" revision is not
listed but contained.
I'm not sure if "all()" should filter out "null", but "all()" is stated as
'the same as "0:tip"' (except that it doesn't reorder the subset, I think.)
This patch is intended to avoid exposing a fullreposet to graphmod.dagwalker(),
which would result in strange drawing in future version:
|
o changeset: 0:f8035bb17114
| user: test
| date: Thu Jan 01 00:00:00 1970 +0000
| summary: add a
caused by:
parents = sorted(set([p.rev() for p in ctx.parents()
if p.rev() in revs]))
We cannot add "and p.rev() != nullrev" here because revs may actually include
"null" revision.
If a user has a revsetalias defined, it is their explicit wish for
this alias to be parsed as a revset and nothing else. Although the
case of the alias being short enough and only contain the letters a-f
is probably kind of rare, it may still happen.
