$ HGENCODING=utf-8 $ export HGENCODING $ cat > testrevset.py << EOF > import mercurial.revset > > baseset = mercurial.revset.baseset > > def r3232(repo, subset, x): > """"simple revset that return [3,2,3,2] > > revisions duplicated on purpose. > """ > if 3 not in subset: > if 2 in subset: > return baseset([2, 2]) > return baseset() > return baseset([3, 3, 2, 2]) > > mercurial.revset.symbols[b'r3232'] = r3232 > EOF $ cat >> $HGRCPATH << EOF > [extensions] > drawdag=$TESTDIR/drawdag.py > testrevset=$TESTTMP/testrevset.py > EOF $ try() { > hg debugrevspec --debug "$@" > } $ log() { > hg log --template '{rev}\n' -r "$1" > } extension to build '_intlist()' and '_hexlist()', which is necessary because these predicates use '\0' as a separator: $ cat < debugrevlistspec.py > from mercurial import ( > node as nodemod, > registrar, > revset, > revsetlang, > ) > from mercurial.utils import stringutil > cmdtable = {} > command = registrar.command(cmdtable) > @command(b'debugrevlistspec', > [(b'', b'optimize', None, b'print parsed tree after optimizing'), > (b'', b'bin', None, b'unhexlify arguments')]) > def debugrevlistspec(ui, repo, fmt, *args, **opts): > if opts['bin']: > args = map(nodemod.bin, args) > expr = revsetlang.formatspec(fmt, list(args)) > if ui.verbose: > tree = revsetlang.parse(expr, lookup=revset.lookupfn(repo)) > ui.note(revsetlang.prettyformat(tree), b"\n") > if opts["optimize"]: > opttree = revsetlang.optimize(revsetlang.analyze(tree)) > ui.note(b"* optimized:\n", revsetlang.prettyformat(opttree), > b"\n") > func = revset.match(ui, expr, lookup=revset.lookupfn(repo)) > revs = func(repo) > if ui.verbose: > ui.note(b"* set:\n", stringutil.prettyrepr(revs), b"\n") > for c in revs: > ui.write(b"%d\n" % c) > EOF $ cat <> $HGRCPATH > [extensions] > debugrevlistspec = $TESTTMP/debugrevlistspec.py > EOF $ trylist() { > hg debugrevlistspec --debug "$@" > } $ hg init repo $ cd repo $ echo a > a $ hg branch a marked working directory as branch a (branches are permanent and global, did you want a bookmark?) $ hg ci -Aqm0 $ echo b > b $ hg branch b marked working directory as branch b $ hg ci -Aqm1 $ rm a $ hg branch a-b-c- marked working directory as branch a-b-c- $ hg ci -Aqm2 -u Bob $ hg log -r "extra('branch', 'a-b-c-')" --template '{rev}\n' 2 $ hg log -r "extra('branch')" --template '{rev}\n' 0 1 2 $ hg log -r "extra('branch', 're:a')" --template '{rev} {branch}\n' 0 a 2 a-b-c- $ hg co 1 1 files updated, 0 files merged, 0 files removed, 0 files unresolved $ hg branch +a+b+c+ marked working directory as branch +a+b+c+ $ hg ci -Aqm3 $ hg co 2 # interleave 0 files updated, 0 files merged, 1 files removed, 0 files unresolved $ echo bb > b $ hg branch -- -a-b-c- marked working directory as branch -a-b-c- $ hg ci -Aqm4 -d "May 12 2005" $ hg co 3 2 files updated, 0 files merged, 0 files removed, 0 files unresolved $ hg branch !a/b/c/ marked working directory as branch !a/b/c/ $ hg ci -Aqm"5 bug" $ hg merge 4 1 files updated, 0 files merged, 1 files removed, 0 files unresolved (branch merge, don't forget to commit) $ hg branch _a_b_c_ marked working directory as branch _a_b_c_ $ hg ci -Aqm"6 issue619" $ hg branch .a.b.c. marked working directory as branch .a.b.c. $ hg ci -Aqm7 $ hg branch all marked working directory as branch all $ hg co 4 0 files updated, 0 files merged, 0 files removed, 0 files unresolved $ hg branch é marked working directory as branch \xc3\xa9 (esc) $ hg ci -Aqm9 $ hg tag -r6 1.0 $ hg bookmark -r6 xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx $ hg clone --quiet -U -r 7 . ../remote1 $ hg clone --quiet -U -r 8 . ../remote2 $ echo "[paths]" >> .hg/hgrc $ echo "default = ../remote1" >> .hg/hgrc trivial $ try 0:1 (range (symbol '0') (symbol '1')) * set: 0 1 $ try --optimize : (rangeall None) * optimized: (rangeall None) * set: 0 1 2 3 4 5 6 7 8 9 $ try 3::6 (dagrange (symbol '3') (symbol '6')) * set: 3 5 6 $ try '0|1|2' (or (list (symbol '0') (symbol '1') (symbol '2'))) * set: 0 1 2 names that should work without quoting $ try a (symbol 'a') * set: 0 $ try b-a (minus (symbol 'b') (symbol 'a')) * set: , >> 1 $ try _a_b_c_ (symbol '_a_b_c_') * set: 6 $ try _a_b_c_-a (minus (symbol '_a_b_c_') (symbol 'a')) * set: , >> 6 $ try .a.b.c. (symbol '.a.b.c.') * set: 7 $ try .a.b.c.-a (minus (symbol '.a.b.c.') (symbol 'a')) * set: , >> 7 names that should be caught by fallback mechanism $ try -- '-a-b-c-' (symbol '-a-b-c-') * set: 4 $ log -a-b-c- 4 $ try '+a+b+c+' (symbol '+a+b+c+') * set: 3 $ try '+a+b+c+:' (rangepost (symbol '+a+b+c+')) * set: 3 4 5 6 7 8 9 $ try ':+a+b+c+' (rangepre (symbol '+a+b+c+')) * set: 0 1 2 3 $ try -- '-a-b-c-:+a+b+c+' (range (symbol '-a-b-c-') (symbol '+a+b+c+')) * set: 4 3 $ log '-a-b-c-:+a+b+c+' 4 3 $ try -- -a-b-c--a # complains (minus (minus (minus (negate (symbol 'a')) (symbol 'b')) (symbol 'c')) (negate (symbol 'a'))) abort: unknown revision '-a' [10] $ try é (symbol '\xc3\xa9') * set: 9 no quoting needed $ log ::a-b-c- 0 1 2 quoting needed $ try '"-a-b-c-"-a' (minus (string '-a-b-c-') (symbol 'a')) * set: , >> 4 $ log '1 or 2' 1 2 $ log '1|2' 1 2 $ log '1 and 2' $ log '1&2' $ try '1&2|3' # precedence - and is higher (or (list (and (symbol '1') (symbol '2')) (symbol '3'))) * set: , > 3 $ try '1|2&3' (or (list (symbol '1') (and (symbol '2') (symbol '3')))) * set: , > 1 $ try '1&2&3' # associativity (and (and (symbol '1') (symbol '2')) (symbol '3')) * set: $ try '1|(2|3)' (or (list (symbol '1') (group (or (list (symbol '2') (symbol '3')))))) * set: , > 1 2 3 $ log '1.0' # tag 6 $ log 'a' # branch 0 $ log '2785f51ee' 0 $ log 'date(2005)' 4 $ log 'date(this is a test)' hg: parse error at 10: unexpected token: symbol (date(this is a test) ^ here) [10] $ log 'date()' hg: parse error: date requires a string [10] $ log 'date' abort: unknown revision 'date' [10] $ log 'date(' hg: parse error at 5: not a prefix: end (date( ^ here) [10] $ log 'date("\xy")' hg: parse error: invalid \x escape* (glob) [10] $ log 'date(tip)' hg: parse error: invalid date: 'tip' [10] $ log '0:date' abort: unknown revision 'date' [10] $ log '::"date"' abort: unknown revision 'date' [10] $ hg book date -r 4 $ log '0:date' 0 1 2 3 4 $ log '::date' 0 1 2 4 $ log '::"date"' 0 1 2 4 $ log 'date(2005) and 1::' 4 $ hg book -d date function name should be a symbol $ log '"date"(2005)' hg: parse error: not a symbol [10] keyword arguments $ log 'extra(branch, value=a)' 0 $ log 'extra(branch, a, b)' hg: parse error: extra takes at most 2 positional arguments [10] $ log 'extra(a, label=b)' hg: parse error: extra got multiple values for keyword argument 'label' [10] $ log 'extra(label=branch, default)' hg: parse error: extra got an invalid argument [10] $ log 'extra(branch, foo+bar=baz)' hg: parse error: extra got an invalid argument [10] $ log 'extra(unknown=branch)' hg: parse error: extra got an unexpected keyword argument 'unknown' [10] $ log 'extra((), x)' hg: parse error: first argument to extra must be a string [10] $ log 'extra(label=x, ())' hg: parse error: extra got an invalid argument [10] $ try 'foo=bar|baz' (keyvalue (symbol 'foo') (or (list (symbol 'bar') (symbol 'baz')))) hg: parse error: can't use a key-value pair in this context [10] right-hand side should be optimized recursively $ try --optimize 'foo=(not public())' (keyvalue (symbol 'foo') (group (not (func (symbol 'public') None)))) * optimized: (keyvalue (symbol 'foo') (func (symbol '_notpublic') None)) hg: parse error: can't use a key-value pair in this context [10] relation-subscript operator has the highest binding strength (as function call): $ hg debugrevspec -p parsed 'tip:tip^#generations[-1]' * parsed: (range (symbol 'tip') (relsubscript (parentpost (symbol 'tip')) (symbol 'generations') (negate (symbol '1')))) 9 8 7 6 5 4 $ hg debugrevspec -p parsed --no-show-revs 'not public()#generations[0]' * parsed: (not (relsubscript (func (symbol 'public') None) (symbol 'generations') (symbol '0'))) left-hand side of relation-subscript operator should be optimized recursively: $ hg debugrevspec -p analyzed -p optimized --no-show-revs \ > '(not public())#generations[0]' * analyzed: (relsubscript (not (func (symbol 'public') None)) (symbol 'generations') (symbol '0')) * optimized: (relsubscript (func (symbol '_notpublic') None) (symbol 'generations') (symbol '0')) resolution of subscript and relation-subscript ternary operators: $ hg debugrevspec -p analyzed 'tip[0]' * analyzed: (subscript (symbol 'tip') (symbol '0')) hg: parse error: can't use a subscript in this context [10] $ hg debugrevspec -p analyzed 'tip#rel[0]' * analyzed: (relsubscript (symbol 'tip') (symbol 'rel') (symbol '0')) hg: parse error: unknown identifier: rel [10] $ hg debugrevspec -p analyzed '(tip#rel)[0]' * analyzed: (subscript (relation (symbol 'tip') (symbol 'rel')) (symbol '0')) hg: parse error: can't use a subscript in this context [10] $ hg debugrevspec -p analyzed 'tip#rel[0][1]' * analyzed: (subscript (relsubscript (symbol 'tip') (symbol 'rel') (symbol '0')) (symbol '1')) hg: parse error: can't use a subscript in this context [10] $ hg debugrevspec -p analyzed 'tip#rel0#rel1[1]' * analyzed: (relsubscript (relation (symbol 'tip') (symbol 'rel0')) (symbol 'rel1') (symbol '1')) hg: parse error: unknown identifier: rel1 [10] $ hg debugrevspec -p analyzed 'tip#rel0[0]#rel1[1]' * analyzed: (relsubscript (relsubscript (symbol 'tip') (symbol 'rel0') (symbol '0')) (symbol 'rel1') (symbol '1')) hg: parse error: unknown identifier: rel1 [10] parse errors of relation, subscript and relation-subscript operators: $ hg debugrevspec '[0]' hg: parse error at 0: not a prefix: [ ([0] ^ here) [10] $ hg debugrevspec '.#' hg: parse error at 2: not a prefix: end (.# ^ here) [10] $ hg debugrevspec '#rel' hg: parse error at 0: not a prefix: # (#rel ^ here) [10] $ hg debugrevspec '.#rel[0' hg: parse error at 7: unexpected token: end (.#rel[0 ^ here) [10] $ hg debugrevspec '.]' hg: parse error at 1: invalid token (.] ^ here) [10] $ hg debugrevspec '.#generations[a]' hg: parse error: relation subscript must be an integer or a range [10] $ hg debugrevspec '.#generations[1-2]' hg: parse error: relation subscript must be an integer or a range [10] $ hg debugrevspec '.#generations[foo:bar]' hg: parse error: relation subscript bounds must be integers [10] suggested relations $ hg debugrevspec '.#generafions[0]' hg: parse error: unknown identifier: generafions (did you mean generations?) [10] $ hg debugrevspec '.#f[0]' hg: parse error: unknown identifier: f [10] parsed tree at stages: $ hg debugrevspec -p all '()' * parsed: (group None) * expanded: (group None) * concatenated: (group None) * analyzed: None * optimized: None hg: parse error: missing argument [10] $ hg debugrevspec --no-optimized -p all '()' * parsed: (group None) * expanded: (group None) * concatenated: (group None) * analyzed: None hg: parse error: missing argument [10] $ hg debugrevspec -p parsed -p analyzed -p optimized '(0|1)-1' * parsed: (minus (group (or (list (symbol '0') (symbol '1')))) (symbol '1')) * analyzed: (and (or (list (symbol '0') (symbol '1'))) (not (symbol '1'))) * optimized: (difference (func (symbol '_list') (string '0\x001')) (symbol '1')) 0 $ hg debugrevspec -p unknown '0' abort: invalid stage name: unknown [255] $ hg debugrevspec -p all --optimize '0' abort: cannot use --optimize with --show-stage [255] verify optimized tree: $ hg debugrevspec --verify '0|1' $ hg debugrevspec --verify -v -p analyzed -p optimized 'r3232() & 2' * analyzed: (and (func (symbol 'r3232') None) (symbol '2')) * optimized: (andsmally (func (symbol 'r3232') None) (symbol '2')) * analyzed set: * optimized set: --- analyzed +++ optimized 2 +2 [1] $ hg debugrevspec --no-optimized --verify-optimized '0' abort: cannot use --verify-optimized with --no-optimized [255] Test that symbols only get parsed as functions if there's an opening parenthesis. $ hg book only -r 9 $ log 'only(only)' # Outer "only" is a function, inner "only" is the bookmark 8 9 ':y' behaves like '0:y', but can't be rewritten as such since the revision '0' may be hidden (issue5385) $ try -p parsed -p analyzed ':' * parsed: (rangeall None) * analyzed: (rangeall None) * set: 0 1 2 3 4 5 6 7 8 9 $ try -p analyzed ':1' * analyzed: (rangepre (symbol '1')) * set: 0 1 $ try -p analyzed ':(1|2)' * analyzed: (rangepre (or (list (symbol '1') (symbol '2')))) * set: 0 1 2 $ try -p analyzed ':(1&2)' * analyzed: (rangepre (and (symbol '1') (symbol '2'))) * set: infix/suffix resolution of ^ operator (issue2884, issue5764): x^:y means (x^):y $ try '1^:2' (range (parentpost (symbol '1')) (symbol '2')) * set: 0 1 2 $ try '1^::2' (dagrange (parentpost (symbol '1')) (symbol '2')) * set: 0 1 2 $ try '1^..2' (dagrange (parentpost (symbol '1')) (symbol '2')) * set: 0 1 2 $ try '9^:' (rangepost (parentpost (symbol '9'))) * set: 8 9 $ try '9^::' (dagrangepost (parentpost (symbol '9'))) * set: 8 9 $ try '9^..' (dagrangepost (parentpost (symbol '9'))) * set: 8 9 x^:y should be resolved before omitting group operators $ try '1^(:2)' (parent (symbol '1') (group (rangepre (symbol '2')))) hg: parse error: ^ expects a number 0, 1, or 2 [10] x^:y should be resolved recursively $ try 'sort(1^:2)' (func (symbol 'sort') (range (parentpost (symbol '1')) (symbol '2'))) * set: 0 1 2 $ try '(3^:4)^:2' (range (parentpost (group (range (parentpost (symbol '3')) (symbol '4')))) (symbol '2')) * set: 0 1 2 $ try '(3^::4)^::2' (dagrange (parentpost (group (dagrange (parentpost (symbol '3')) (symbol '4')))) (symbol '2')) * set: 0 1 2 $ try '(9^:)^:' (rangepost (parentpost (group (rangepost (parentpost (symbol '9')))))) * set: 4 5 6 7 8 9 x^ in alias should also be resolved $ try 'A' --config 'revsetalias.A=1^:2' (symbol 'A') * expanded: (range (parentpost (symbol '1')) (symbol '2')) * set: 0 1 2 $ try 'A:2' --config 'revsetalias.A=1^' (range (symbol 'A') (symbol '2')) * expanded: (range (parentpost (symbol '1')) (symbol '2')) * set: 0 1 2 but not beyond the boundary of alias expansion, because the resolution should be made at the parsing stage $ try '1^A' --config 'revsetalias.A=:2' (parent (symbol '1') (symbol 'A')) * expanded: (parent (symbol '1') (rangepre (symbol '2'))) hg: parse error: ^ expects a number 0, 1, or 2 [10] '::' itself isn't a valid expression $ try '::' (dagrangeall None) hg: parse error: can't use '::' in this context [10] ancestor can accept 0 or more arguments $ log 'ancestor()' $ log 'ancestor(1)' 1 $ log 'ancestor(4,5)' 1 $ log 'ancestor(4,5) and 4' $ log 'ancestor(0,0,1,3)' 0 $ log 'ancestor(3,1,5,3,5,1)' 1 $ log 'ancestor(0,1,3,5)' 0 $ log 'ancestor(1,2,3,4,5)' 1 test ancestors $ hg log -G -T '{rev}\n' --config experimental.graphshorten=True @ 9 o 8 | o 7 | o 6 |/| | o 5 o | 4 | o 3 o | 2 |/ o 1 o 0 $ log 'ancestors(5)' 0 1 3 5 $ log 'ancestor(ancestors(5))' 0 $ log '::r3232()' 0 1 2 3 test common ancestors $ hg log -T '{rev}\n' -r 'commonancestors(7 + 9)' 0 1 2 4 $ hg log -T '{rev}\n' -r 'commonancestors(heads(all()))' 0 1 2 4 $ hg log -T '{rev}\n' -r 'commonancestors(9)' 0 1 2 4 8 9 $ hg log -T '{rev}\n' -r 'commonancestors(8 + 9)' 0 1 2 4 8 test the specialized implementation of heads(commonancestors(..)) (2 gcas is tested in test-merge-criss-cross.t) $ hg log -T '{rev}\n' -r 'heads(commonancestors(7 + 9))' 4 $ hg log -T '{rev}\n' -r 'heads(commonancestors(heads(all())))' 4 $ hg log -T '{rev}\n' -r 'heads(commonancestors(9))' 9 $ hg log -T '{rev}\n' -r 'heads(commonancestors(8 + 9))' 8 test ancestor variants of empty revision $ log 'ancestor(none())' $ log 'ancestors(none())' $ log 'commonancestors(none())' $ log 'heads(commonancestors(none()))' test ancestors with depth limit (depth=0 selects the node itself) $ log 'reverse(ancestors(9, depth=0))' 9 (interleaved: '4' would be missing if heap queue were higher depth first) $ log 'reverse(ancestors(8:9, depth=1))' 9 8 4 (interleaved: '2' would be missing if heap queue were higher depth first) $ log 'reverse(ancestors(7+8, depth=2))' 8 7 6 5 4 2 (walk example above by separate queries) $ log 'reverse(ancestors(8, depth=2)) + reverse(ancestors(7, depth=2))' 8 4 2 7 6 5 (walk 2nd and 3rd ancestors) $ log 'reverse(ancestors(7, depth=3, startdepth=2))' 5 4 3 2 (interleaved: '4' would be missing if higher-depth ancestors weren't scanned) $ log 'reverse(ancestors(7+8, depth=2, startdepth=2))' 5 4 2 (note that 'ancestors(x, depth=y, startdepth=z)' does not identical to 'ancestors(x, depth=y) - ancestors(x, depth=z-1)' because a node may have multiple depths) $ log 'reverse(ancestors(7+8, depth=2) - ancestors(7+8, depth=1))' 5 2 test bad arguments passed to ancestors() $ log 'ancestors(., depth=-1)' hg: parse error: negative depth [10] $ log 'ancestors(., depth=foo)' hg: parse error: ancestors expects an integer depth [10] test descendants $ hg log -G -T '{rev}\n' --config experimental.graphshorten=True @ 9 o 8 | o 7 | o 6 |/| | o 5 o | 4 | o 3 o | 2 |/ o 1 o 0 (null is ultimate root and has optimized path) $ log 'null:4 & descendants(null)' -1 0 1 2 3 4 (including merge) $ log ':8 & descendants(2)' 2 4 6 7 8 (multiple roots) $ log ':8 & descendants(2+5)' 2 4 5 6 7 8 test descendants with depth limit (depth=0 selects the node itself) $ log 'descendants(0, depth=0)' 0 $ log 'null: & descendants(null, depth=0)' -1 (p2 = null should be ignored) $ log 'null: & descendants(null, depth=2)' -1 0 1 (multiple paths: depth(6) = (2, 3)) $ log 'descendants(1+3, depth=2)' 1 2 3 4 5 6 (multiple paths: depth(5) = (1, 2), depth(6) = (2, 3)) $ log 'descendants(3+1, depth=2, startdepth=2)' 4 5 6 (multiple depths: depth(6) = (0, 2, 4), search for depth=2) $ log 'descendants(0+3+6, depth=3, startdepth=1)' 1 2 3 4 5 6 7 (multiple depths: depth(6) = (0, 4), no match) $ log 'descendants(0+6, depth=3, startdepth=1)' 1 2 3 4 5 7 test ancestors/descendants relation: $ log 'tip#generations' 0 1 2 4 8 9 $ log '3#g' 0 1 3 5 6 7 $ hg debugrevspec -p parsed 'tip#g' * parsed: (relation (symbol 'tip') (symbol 'g')) 0 1 2 4 8 9 test ancestors/descendants relation subscript: $ log 'tip#generations[0]' 9 $ log '.#generations[-1]' 8 $ log '.#g[(-1)]' 8 $ log '6#generations[0:1]' 6 7 $ log '6#generations[-1:1]' 4 5 6 7 $ log '6#generations[0:]' 6 7 $ log '5#generations[:0]' 0 1 3 5 $ log '3#generations[:]' 0 1 3 5 6 7 $ log 'tip#generations[1:-1]' $ hg debugrevspec -p parsed 'roots(:)#g[2]' * parsed: (relsubscript (func (symbol 'roots') (rangeall None)) (symbol 'g') (symbol '2')) 2 3 test author $ log 'author(bob)' 2 $ log 'author("re:bob|test")' 0 1 2 3 4 5 6 7 8 9 $ log 'author(r"re:\S")' 0 1 2 3 4 5 6 7 8 9 $ log 'branch(é)' 8 9 $ log 'branch(a)' 0 $ hg log -r 'branch("re:a")' --template '{rev} {branch}\n' 0 a 2 a-b-c- 3 +a+b+c+ 4 -a-b-c- 5 !a/b/c/ 6 _a_b_c_ 7 .a.b.c. $ log 'children(ancestor(4,5))' 2 3 $ log 'children(4)' 6 8 $ log 'children(null)' 0 $ log 'closed()' $ log 'contains(a)' 0 1 3 5 $ log 'contains("../repo/a")' 0 1 3 5 $ log 'desc(B)' 5 $ hg log -r 'desc(r"re:S?u")' --template "{rev} {desc|firstline}\n" 5 5 bug 6 6 issue619 $ log 'descendants(2 or 3)' 2 3 4 5 6 7 8 9 $ log 'file("b*")' 1 4 $ log 'filelog("b")' 1 4 $ log 'filelog("../repo/b")' 1 4 $ log 'follow()' 0 1 2 4 8 9 $ log 'grep("issue\d+")' 6 $ try 'grep("(")' # invalid regular expression (func (symbol 'grep') (string '(')) hg: parse error: invalid match pattern: (unbalanced parenthesis|missing \),.*) (re) [10] $ log 'desc("re:(")' hg: parse error: invalid regular expression: (unbalanced parenthesis|missing \),.*) (re) [10] $ try 'grep("\bissue\d+")' (func (symbol 'grep') (string '\x08issue\\d+')) * set: , > $ try 'grep(r"\bissue\d+")' (func (symbol 'grep') (string '\\bissue\\d+')) * set: , > 6 $ try 'grep(r"\")' hg: parse error at 7: unterminated string (grep(r"\") ^ here) [10] $ log 'head()' 0 1 2 3 4 5 6 7 9 Test heads $ log 'heads(6::)' 7 heads() can be computed in subset '9:' $ hg debugrevspec -s '9: & heads(all())' * set: , > 9 but should follow the order of the subset $ log 'heads(all())' 7 9 $ log 'heads(tip:0)' 7 9 $ log 'tip:0 & heads(all())' 9 7 $ log 'tip:0 & heads(0:tip)' 9 7 $ log 'keyword(issue)' 6 $ log 'keyword("test a")' Test first (=limit) and last $ log 'limit(head(), 1)' 0 $ log 'limit(author("re:bob|test"), 3, 5)' 5 6 7 $ log 'limit(author("re:bob|test"), offset=6)' 6 $ log 'limit(author("re:bob|test"), offset=10)' $ log 'limit(all(), 1, -1)' hg: parse error: negative offset [10] $ log 'limit(all(), -1)' hg: parse error: negative number to select [10] $ log 'limit(all(), 0)' $ log 'last(all(), -1)' hg: parse error: negative number to select [10] $ log 'last(all(), 0)' $ log 'last(all(), 1)' 9 $ log 'last(all(), 2)' 8 9 Test smartset.slice() by first/last() (using unoptimized set, filteredset as example) $ hg debugrevspec --no-show-revs -s '0:7 & branch("re:")' * set: , > $ log 'limit(0:7 & branch("re:"), 3, 4)' 4 5 6 $ log 'limit(7:0 & branch("re:"), 3, 4)' 3 2 1 $ log 'last(0:7 & branch("re:"), 2)' 6 7 (using baseset) $ hg debugrevspec --no-show-revs -s 0+1+2+3+4+5+6+7 * set: $ hg debugrevspec --no-show-revs -s 0::7 * set: $ log 'limit(0+1+2+3+4+5+6+7, 3, 4)' 4 5 6 $ log 'limit(sort(0::7, rev), 3, 4)' 4 5 6 $ log 'limit(sort(0::7, -rev), 3, 4)' 3 2 1 $ log 'last(sort(0::7, rev), 2)' 6 7 $ hg debugrevspec -s 'limit(sort(0::7, rev), 3, 6)' * set: 6 7 $ hg debugrevspec -s 'limit(sort(0::7, rev), 3, 9)' * set: $ hg debugrevspec -s 'limit(sort(0::7, -rev), 3, 6)' * set: 1 0 $ hg debugrevspec -s 'limit(sort(0::7, -rev), 3, 9)' * set: $ hg debugrevspec -s 'limit(0::7, 0)' * set: (using spanset) $ hg debugrevspec --no-show-revs -s 0:7 * set: $ log 'limit(0:7, 3, 4)' 4 5 6 $ log 'limit(7:0, 3, 4)' 3 2 1 $ log 'limit(0:7, 3, 6)' 6 7 $ log 'limit(7:0, 3, 6)' 1 0 $ log 'last(0:7, 2)' 6 7 $ hg debugrevspec -s 'limit(0:7, 3, 6)' * set: 6 7 $ hg debugrevspec -s 'limit(0:7, 3, 9)' * set: $ hg debugrevspec -s 'limit(7:0, 3, 6)' * set: 1 0 $ hg debugrevspec -s 'limit(7:0, 3, 9)' * set: $ hg debugrevspec -s 'limit(0:7, 0)' * set: Test order of first/last revisions $ hg debugrevspec -s 'first(4:0, 3) & 3:' * set: , > 4 3 $ hg debugrevspec -s '3: & first(4:0, 3)' * set: , > 3 4 $ hg debugrevspec -s 'last(4:0, 3) & :1' * set: , > 1 0 $ hg debugrevspec -s ':1 & last(4:0, 3)' * set: , > 0 1 Test scmutil.revsingle() should return the last revision $ hg debugrevspec -s 'last(0::)' * set: > 9 $ hg identify -r '0::' --num 9 Test matching $ log 'matching(6)' 6 $ log 'matching(6:7, "phase parents user date branch summary files description substate")' 6 7 Testing min and max max: simple $ log 'max(contains(a))' 5 max: simple on unordered set) $ log 'max((4+0+2+5+7) and contains(a))' 5 max: no result $ log 'max(contains(stringthatdoesnotappearanywhere))' max: no result on unordered set $ log 'max((4+0+2+5+7) and contains(stringthatdoesnotappearanywhere))' min: simple $ log 'min(contains(a))' 0 min: simple on unordered set $ log 'min((4+0+2+5+7) and contains(a))' 0 min: empty $ log 'min(contains(stringthatdoesnotappearanywhere))' min: empty on unordered set $ log 'min((4+0+2+5+7) and contains(stringthatdoesnotappearanywhere))' $ log 'merge()' 6 $ log 'branchpoint()' 1 4 $ log 'modifies(b)' 4 $ log 'modifies("path:b")' 4 $ log 'modifies("*")' 4 $ log 'modifies("set:modified()")' 4 $ log 'id(5)' $ log 'only(9)' 8 9 $ log 'only(8)' 8 $ log 'only(9, 5)' 2 4 8 9 $ log 'only(7 + 9, 5 + 2)' 4 6 7 8 9 Test empty set input $ log 'only(p2())' $ log 'only(p1(), p2())' 0 1 2 4 8 9 Test '%' operator $ log '9%' 8 9 $ log '9%5' 2 4 8 9 $ log '(7 + 9)%(5 + 2)' 4 6 7 8 9 Test operand of '%' is optimized recursively (issue4670) $ try --optimize '8:9-8%' (onlypost (minus (range (symbol '8') (symbol '9')) (symbol '8'))) * optimized: (func (symbol 'only') (difference (range (symbol '8') (symbol '9')) (symbol '8'))) * set: 8 9 $ try --optimize '(9)%(5)' (only (group (symbol '9')) (group (symbol '5'))) * optimized: (func (symbol 'only') (list (symbol '9') (symbol '5'))) * set: 2 4 8 9 Test the order of operations $ log '7 + 9%5 + 2' 7 2 4 8 9 Test explicit numeric revision $ log 'rev(-2)' $ log 'rev(-1)' -1 $ log 'rev(0)' 0 $ log 'rev(9)' 9 $ log 'rev(10)' $ log 'rev(tip)' hg: parse error: rev expects a number [10] Test hexadecimal revision $ log 'id(2)' $ log 'id(8)' 3 $ hg log --template '{rev}\n' -r 'id(x8)' 3 $ hg log --template '{rev}\n' -r 'x8' 3 $ hg log --template '{rev}\n' -r 'id(x)' $ hg log --template '{rev}\n' -r 'x' abort: ambiguous revision identifier: x [10] $ log 'id(23268)' 4 $ log 'id(2785f51eece)' 0 $ log 'id(d5d0dcbdc4d9ff5dbb2d336f32f0bb561c1a532c)' 8 $ log 'id(d5d0dcbdc4a)' $ log 'id(d5d0dcbdc4w)' $ log 'id(d5d0dcbdc4d9ff5dbb2d336f32f0bb561c1a532d)' $ log 'id(d5d0dcbdc4d9ff5dbb2d336f32f0bb561c1a532q)' $ log 'id(1.0)' $ log 'id(xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx)' Test null revision $ log '(null)' -1 $ log '(null:0)' -1 0 $ log '(0:null)' 0 -1 $ log 'null::0' -1 0 $ log 'null:tip - 0:' -1 $ log 'null: and null::' | head -1 -1 $ log 'null: or 0:' | head -2 -1 0 $ log 'ancestors(null)' -1 $ log 'reverse(null:)' | tail -2 0 -1 $ log 'first(null:)' -1 $ log 'min(null:)' BROKEN: should be '-1' $ log 'tip:null and all()' | tail -2 1 0 Test working-directory revision $ hg debugrevspec 'wdir()' 2147483647 $ hg debugrevspec 'wdir()^' 9 $ hg up 7 0 files updated, 0 files merged, 1 files removed, 0 files unresolved $ hg debugrevspec 'wdir()^' 7 $ hg debugrevspec 'wdir()^0' 2147483647 $ hg debugrevspec 'wdir()~3' 5 $ hg debugrevspec 'ancestors(wdir())' 0 1 2 3 4 5 6 7 2147483647 $ hg debugrevspec '0:wdir() & ancestor(wdir())' 2147483647 $ hg debugrevspec '0:wdir() & ancestor(.:wdir())' 4 $ hg debugrevspec '0:wdir() & ancestor(wdir(), wdir())' 2147483647 $ hg debugrevspec '0:wdir() & ancestor(wdir(), tip)' 4 $ hg debugrevspec 'null:wdir() & ancestor(wdir(), null)' -1 $ hg debugrevspec 'wdir()~0' 2147483647 $ hg debugrevspec 'p1(wdir())' 7 $ hg debugrevspec 'p2(wdir())' $ hg debugrevspec 'parents(wdir())' 7 $ hg debugrevspec 'wdir()^1' 7 $ hg debugrevspec 'wdir()^2' $ hg debugrevspec 'wdir()^3' hg: parse error: ^ expects a number 0, 1, or 2 [10] For tests consistency $ hg up 9 1 files updated, 0 files merged, 0 files removed, 0 files unresolved $ hg debugrevspec 'tip or wdir()' 9 2147483647 $ hg debugrevspec '0:tip and wdir()' $ log '0:wdir()' | tail -3 8 9 2147483647 $ log 'wdir():0' | head -3 2147483647 9 8 $ log 'wdir():wdir()' 2147483647 $ log '(all() + wdir()) & min(. + wdir())' 9 $ log '(all() + wdir()) & max(. + wdir())' 2147483647 $ log 'first(wdir() + .)' 2147483647 $ log 'last(. + wdir())' 2147483647 Test working-directory integer revision and node id $ hg debugrevspec '2147483647' 2147483647 $ hg debugrevspec 'rev(2147483647)' 2147483647 $ hg debugrevspec 'ffffffffffffffffffffffffffffffffffffffff' 2147483647 $ hg debugrevspec 'ffffffffffff' 2147483647 $ hg debugrevspec 'id(ffffffffffffffffffffffffffffffffffffffff)' 2147483647 $ hg debugrevspec 'id(ffffffffffff)' 2147483647 $ hg debugrevspec 'ffffffffffff+000000000000' 2147483647 -1 $ cd .. Test short 'ff...' hash collision $ hg init wdir-hashcollision $ cd wdir-hashcollision $ cat <> .hg/hgrc > [experimental] > evolution.createmarkers=True > EOF $ echo 0 > a $ hg ci -qAm 0 $ for i in 2463 2961 6726 78127; do > hg up -q 0 > echo $i > a > hg ci -qm $i > done $ hg up -q null $ hg log -r '0:wdir()' -T '{rev}:{node} {shortest(node, 3)}\n' 0:b4e73ffab476aa0ee32ed81ca51e07169844bc6a b4e 1:fffbae3886c8fbb2114296380d276fd37715d571 fffba 2:fffb6093b00943f91034b9bdad069402c834e572 fffb6 3:fff48a9b9de34a4d64120c29548214c67980ade3 fff4 4:ffff85cff0ff78504fcdc3c0bc10de0c65379249 ffff8 2147483647:ffffffffffffffffffffffffffffffffffffffff fffff $ hg debugobsolete fffbae3886c8fbb2114296380d276fd37715d571 1 new obsolescence markers obsoleted 1 changesets $ hg debugrevspec 'fff' abort: ambiguous revision identifier: fff [10] $ hg debugrevspec 'ffff' abort: ambiguous revision identifier: ffff [10] $ hg debugrevspec 'fffb' abort: ambiguous revision identifier: fffb [10] BROKEN should be '2' (node lookup uses unfiltered repo) $ hg debugrevspec 'id(fffb)' BROKEN should be '2' (node lookup uses unfiltered repo) $ hg debugrevspec 'ffff8' 4 $ hg debugrevspec 'fffff' 2147483647 $ cd .. Test branch() with wdir() $ cd repo $ log '0:wdir() & branch("literal:é")' 8 9 2147483647 $ log '0:wdir() & branch("re:é")' 8 9 2147483647 $ log '0:wdir() & branch("re:^a")' 0 2 $ log '0:wdir() & branch(8)' 8 9 2147483647 branch(wdir()) returns all revisions belonging to the working branch. The wdir itself isn't returned unless it is explicitly populated. $ log 'branch(wdir())' 8 9 $ log '0:wdir() & branch(wdir())' 8 9 2147483647 $ log 'outgoing()' 8 9 $ log 'outgoing("../remote1")' 8 9 $ log 'outgoing("../remote2")' 3 5 6 7 9 $ log 'p1(merge())' 5 $ log 'p2(merge())' 4 $ log 'parents(merge())' 4 5 $ hg merge 7 0 files updated, 0 files merged, 0 files removed, 0 files unresolved (branch merge, don't forget to commit) $ log '0:wdir() & merge()' 6 2147483647 $ hg update -qC . $ log '0:wdir() & merge()' 6 $ log 'p1(branchpoint())' 0 2 $ log 'p2(branchpoint())' $ log 'parents(branchpoint())' 0 2 $ log 'removes(a)' 2 $ log 'roots(all())' 0 $ log 'reverse(2 or 3 or 4 or 5)' 5 4 3 2 $ log 'reverse(all())' 9 8 7 6 5 4 3 2 1 0 $ log 'reverse(all()) & filelog(b)' 4 1 $ log 'rev(5)' 5 $ log 'sort(limit(reverse(all()), 3))' 7 8 9 $ log 'sort(2 or 3 or 4 or 5, date)' 2 3 5 4 $ log 'tagged()' 6 $ log 'tag()' 6 $ log 'tag(1.0)' 6 $ log 'tag(tip)' 9 Test order of revisions in compound expression ---------------------------------------------- The general rule is that only the outermost (= leftmost) predicate can enforce its ordering requirement. The other predicates should take the ordering defined by it. 'A & B' should follow the order of 'A': $ log '2:0 & 0::2' 2 1 0 'head()' combines sets in right order: $ log '2:0 & head()' 2 1 0 'x:y' takes ordering parameter into account: $ try -p optimized '3:0 & 0:3 & not 2:1' * optimized: (difference (and (range (symbol '3') (symbol '0')) (range (symbol '0') (symbol '3'))) (range (symbol '2') (symbol '1'))) * set: , >, >> 3 0 'a + b', which is optimized to '_list(a b)', should take the ordering of the left expression: $ try --optimize '2:0 & (0 + 1 + 2)' (and (range (symbol '2') (symbol '0')) (group (or (list (symbol '0') (symbol '1') (symbol '2'))))) * optimized: (and (range (symbol '2') (symbol '0')) (func (symbol '_list') (string '0\x001\x002'))) * set: , > 2 1 0 'A + B' should take the ordering of the left expression: $ try --optimize '2:0 & (0:1 + 2)' (and (range (symbol '2') (symbol '0')) (group (or (list (range (symbol '0') (symbol '1')) (symbol '2'))))) * optimized: (and (range (symbol '2') (symbol '0')) (or (list (range (symbol '0') (symbol '1')) (symbol '2')))) * set: , , >> 2 1 0 '_intlist(a b)' should behave like 'a + b': $ trylist --optimize '2:0 & %ld' 0 1 2 (and (range (symbol '2') (symbol '0')) (func (symbol '_intlist') (string '0\x001\x002'))) * optimized: (andsmally (range (symbol '2') (symbol '0')) (func (symbol '_intlist') (string '0\x001\x002'))) * set: , > 2 1 0 $ trylist --optimize '%ld & 2:0' 0 2 1 (and (func (symbol '_intlist') (string '0\x002\x001')) (range (symbol '2') (symbol '0'))) * optimized: (and (func (symbol '_intlist') (string '0\x002\x001')) (range (symbol '2') (symbol '0'))) * set: , > 0 2 1 '_hexlist(a b)' should behave like 'a + b': $ trylist --optimize --bin '2:0 & %ln' `hg log -T '{node} ' -r0:2` (and (range (symbol '2') (symbol '0')) (func (symbol '_hexlist') (string '*'))) (glob) * optimized: (and (range (symbol '2') (symbol '0')) (func (symbol '_hexlist') (string '*'))) (glob) * set: , > 2 1 0 $ trylist --optimize --bin '%ln & 2:0' `hg log -T '{node} ' -r0+2+1` (and (func (symbol '_hexlist') (string '*')) (glob) (range (symbol '2') (symbol '0'))) * optimized: (andsmally (func (symbol '_hexlist') (string '*')) (glob) (range (symbol '2') (symbol '0'))) * set: 0 2 1 '_list' should not go through the slow follow-order path if order doesn't matter: $ try -p optimized '2:0 & not (0 + 1)' * optimized: (difference (range (symbol '2') (symbol '0')) (func (symbol '_list') (string '0\x001'))) * set: , >> 2 $ try -p optimized '2:0 & not (0:2 & (0 + 1))' * optimized: (difference (range (symbol '2') (symbol '0')) (and (range (symbol '0') (symbol '2')) (func (symbol '_list') (string '0\x001')))) * set: , >> 2 because 'present()' does nothing other than suppressing an error, the ordering requirement should be forwarded to the nested expression $ try -p optimized 'present(2 + 0 + 1)' * optimized: (func (symbol 'present') (func (symbol '_list') (string '2\x000\x001'))) * set: 2 0 1 $ try --optimize '2:0 & present(0 + 1 + 2)' (and (range (symbol '2') (symbol '0')) (func (symbol 'present') (or (list (symbol '0') (symbol '1') (symbol '2'))))) * optimized: (and (range (symbol '2') (symbol '0')) (func (symbol 'present') (func (symbol '_list') (string '0\x001\x002')))) * set: , > 2 1 0 'reverse()' should take effect only if it is the outermost expression: $ try --optimize '0:2 & reverse(all())' (and (range (symbol '0') (symbol '2')) (func (symbol 'reverse') (func (symbol 'all') None))) * optimized: (and (range (symbol '0') (symbol '2')) (func (symbol 'reverse') (func (symbol 'all') None))) * set: , > 0 1 2 'sort()' should take effect only if it is the outermost expression: $ try --optimize '0:2 & sort(all(), -rev)' (and (range (symbol '0') (symbol '2')) (func (symbol 'sort') (list (func (symbol 'all') None) (negate (symbol 'rev'))))) * optimized: (and (range (symbol '0') (symbol '2')) (func (symbol 'sort') (list (func (symbol 'all') None) (string '-rev')))) * set: , > 0 1 2 invalid argument passed to noop sort(): $ log '0:2 & sort()' hg: parse error: sort requires one or two arguments [10] $ log '0:2 & sort(all(), -invalid)' hg: parse error: unknown sort key '-invalid' [10] for 'A & f(B)', 'B' should not be affected by the order of 'A': $ try --optimize '2:0 & first(1 + 0 + 2)' (and (range (symbol '2') (symbol '0')) (func (symbol 'first') (or (list (symbol '1') (symbol '0') (symbol '2'))))) * optimized: (and (range (symbol '2') (symbol '0')) (func (symbol 'first') (func (symbol '_list') (string '1\x000\x002')))) * set: , > 1 $ try --optimize '2:0 & not last(0 + 2 + 1)' (and (range (symbol '2') (symbol '0')) (not (func (symbol 'last') (or (list (symbol '0') (symbol '2') (symbol '1')))))) * optimized: (difference (range (symbol '2') (symbol '0')) (func (symbol 'last') (func (symbol '_list') (string '0\x002\x001')))) * set: , >> 2 0 for 'A & (op)(B)', 'B' should not be affected by the order of 'A': $ try --optimize '2:0 & (1 + 0 + 2):(0 + 2 + 1)' (and (range (symbol '2') (symbol '0')) (range (group (or (list (symbol '1') (symbol '0') (symbol '2')))) (group (or (list (symbol '0') (symbol '2') (symbol '1')))))) * optimized: (and (range (symbol '2') (symbol '0')) (range (func (symbol '_list') (string '1\x000\x002')) (func (symbol '_list') (string '0\x002\x001')))) * set: , > 1 'A & B' can be rewritten as 'flipand(B, A)' by weight. $ try --optimize 'contains("glob:*") & (2 + 0 + 1)' (and (func (symbol 'contains') (string 'glob:*')) (group (or (list (symbol '2') (symbol '0') (symbol '1'))))) * optimized: (andsmally (func (symbol 'contains') (string 'glob:*')) (func (symbol '_list') (string '2\x000\x001'))) * set: , > 0 1 2 and in this example, 'A & B' is rewritten as 'B & A', but 'A' overrides the order appropriately: $ try --optimize 'reverse(contains("glob:*")) & (0 + 2 + 1)' (and (func (symbol 'reverse') (func (symbol 'contains') (string 'glob:*'))) (group (or (list (symbol '0') (symbol '2') (symbol '1'))))) * optimized: (andsmally (func (symbol 'reverse') (func (symbol 'contains') (string 'glob:*'))) (func (symbol '_list') (string '0\x002\x001'))) * set: , > 2 1 0 test sort revset -------------------------------------------- test when adding two unordered revsets $ log 'sort(keyword(issue) or modifies(b))' 4 6 test when sorting a reversed collection in the same way it is $ log 'sort(reverse(all()), -rev)' 9 8 7 6 5 4 3 2 1 0 test when sorting a reversed collection $ log 'sort(reverse(all()), rev)' 0 1 2 3 4 5 6 7 8 9 test sorting two sorted collections in different orders $ log 'sort(outgoing() or reverse(removes(a)), rev)' 2 8 9 test sorting two sorted collections in different orders backwards $ log 'sort(outgoing() or reverse(removes(a)), -rev)' 9 8 2 test empty sort key which is noop $ log 'sort(0 + 2 + 1, "")' 0 2 1 test invalid sort keys $ log 'sort(all(), -invalid)' hg: parse error: unknown sort key '-invalid' [10] $ cd .. test sorting by multiple keys including variable-length strings $ hg init sorting $ cd sorting $ cat <> .hg/hgrc > [ui] > logtemplate = '{rev} {branch|p5}{desc|p5}{author|p5}{date|hgdate}\n' > [templatealias] > p5(s) = pad(s, 5) > EOF $ hg branch -qf b12 $ hg ci -m m111 -u u112 -d '111 10800' $ hg branch -qf b11 $ hg ci -m m12 -u u111 -d '112 7200' $ hg branch -qf b111 $ hg ci -m m11 -u u12 -d '111 3600' $ hg branch -qf b112 $ hg ci -m m111 -u u11 -d '120 0' $ hg branch -qf b111 $ hg ci -m m112 -u u111 -d '110 14400' created new head compare revisions (has fast path): $ hg log -r 'sort(all(), rev)' 0 b12 m111 u112 111 10800 1 b11 m12 u111 112 7200 2 b111 m11 u12 111 3600 3 b112 m111 u11 120 0 4 b111 m112 u111 110 14400 $ hg log -r 'sort(all(), -rev)' 4 b111 m112 u111 110 14400 3 b112 m111 u11 120 0 2 b111 m11 u12 111 3600 1 b11 m12 u111 112 7200 0 b12 m111 u112 111 10800 compare variable-length strings (issue5218): $ hg log -r 'sort(all(), branch)' 1 b11 m12 u111 112 7200 2 b111 m11 u12 111 3600 4 b111 m112 u111 110 14400 3 b112 m111 u11 120 0 0 b12 m111 u112 111 10800 $ hg log -r 'sort(all(), -branch)' 0 b12 m111 u112 111 10800 3 b112 m111 u11 120 0 2 b111 m11 u12 111 3600 4 b111 m112 u111 110 14400 1 b11 m12 u111 112 7200 $ hg log -r 'sort(all(), desc)' 2 b111 m11 u12 111 3600 0 b12 m111 u112 111 10800 3 b112 m111 u11 120 0 4 b111 m112 u111 110 14400 1 b11 m12 u111 112 7200 $ hg log -r 'sort(all(), -desc)' 1 b11 m12 u111 112 7200 4 b111 m112 u111 110 14400 0 b12 m111 u112 111 10800 3 b112 m111 u11 120 0 2 b111 m11 u12 111 3600 $ hg log -r 'sort(all(), user)' 3 b112 m111 u11 120 0 1 b11 m12 u111 112 7200 4 b111 m112 u111 110 14400 0 b12 m111 u112 111 10800 2 b111 m11 u12 111 3600 $ hg log -r 'sort(all(), -user)' 2 b111 m11 u12 111 3600 0 b12 m111 u112 111 10800 1 b11 m12 u111 112 7200 4 b111 m112 u111 110 14400 3 b112 m111 u11 120 0 compare dates (tz offset should have no effect): $ hg log -r 'sort(all(), date)' 4 b111 m112 u111 110 14400 0 b12 m111 u112 111 10800 2 b111 m11 u12 111 3600 1 b11 m12 u111 112 7200 3 b112 m111 u11 120 0 $ hg log -r 'sort(all(), -date)' 3 b112 m111 u11 120 0 1 b11 m12 u111 112 7200 0 b12 m111 u112 111 10800 2 b111 m11 u12 111 3600 4 b111 m112 u111 110 14400 be aware that 'sort(x, -k)' is not exactly the same as 'reverse(sort(x, k))' because '-k' reverses the comparison, not the list itself: $ hg log -r 'sort(0 + 2, date)' 0 b12 m111 u112 111 10800 2 b111 m11 u12 111 3600 $ hg log -r 'sort(0 + 2, -date)' 0 b12 m111 u112 111 10800 2 b111 m11 u12 111 3600 $ hg log -r 'reverse(sort(0 + 2, date))' 2 b111 m11 u12 111 3600 0 b12 m111 u112 111 10800 sort by multiple keys: $ hg log -r 'sort(all(), "branch -rev")' 1 b11 m12 u111 112 7200 4 b111 m112 u111 110 14400 2 b111 m11 u12 111 3600 3 b112 m111 u11 120 0 0 b12 m111 u112 111 10800 $ hg log -r 'sort(all(), "-desc -node")' 1 b11 m12 u111 112 7200 4 b111 m112 u111 110 14400 3 b112 m111 u11 120 0 0 b12 m111 u112 111 10800 2 b111 m11 u12 111 3600 $ hg log -r 'sort(all(), "user -branch date rev")' 3 b112 m111 u11 120 0 4 b111 m112 u111 110 14400 1 b11 m12 u111 112 7200 0 b12 m111 u112 111 10800 2 b111 m11 u12 111 3600 sort including wdir (rev/-rev has fast path): $ hg log -r 'sort(. + wdir(), rev)' -T '{rev}\n' 4 2147483647 $ hg log -r 'sort(. + wdir(), -rev)' -T '{rev}\n' 2147483647 4 $ hg log -r 'sort(. + wdir(), "branch rev")' -T '{rev}\n' 4 2147483647 $ hg log -r 'sort(. + wdir(), "branch -rev")' -T '{rev}\n' 2147483647 4 $ hg log -r 'sort(. + wdir(), node)' -T '{node}\n' ec7c1c90b589ade8603d5fb619dc6c25173a723f ffffffffffffffffffffffffffffffffffffffff $ hg log -r 'sort(. + wdir(), -node)' -T '{node}\n' ffffffffffffffffffffffffffffffffffffffff ec7c1c90b589ade8603d5fb619dc6c25173a723f toposort prioritises graph branches $ hg up 2 0 files updated, 0 files merged, 0 files removed, 0 files unresolved $ touch a $ hg addremove adding a $ hg ci -m 't1' -u 'tu' -d '130 0' created new head $ echo 'a' >> a $ hg ci -m 't2' -u 'tu' -d '130 0' $ hg book book1 $ hg up 4 0 files updated, 0 files merged, 1 files removed, 0 files unresolved (leaving bookmark book1) $ touch a $ hg addremove adding a $ hg ci -m 't3' -u 'tu' -d '130 0' $ hg log -r 'sort(all(), topo)' 7 b111 t3 tu 130 0 4 b111 m112 u111 110 14400 3 b112 m111 u11 120 0 6 b111 t2 tu 130 0 5 b111 t1 tu 130 0 2 b111 m11 u12 111 3600 1 b11 m12 u111 112 7200 0 b12 m111 u112 111 10800 $ hg log -r 'sort(all(), -topo)' 0 b12 m111 u112 111 10800 1 b11 m12 u111 112 7200 2 b111 m11 u12 111 3600 5 b111 t1 tu 130 0 6 b111 t2 tu 130 0 3 b112 m111 u11 120 0 4 b111 m112 u111 110 14400 7 b111 t3 tu 130 0 $ hg log -r 'sort(all(), topo, topo.firstbranch=book1)' 6 b111 t2 tu 130 0 5 b111 t1 tu 130 0 7 b111 t3 tu 130 0 4 b111 m112 u111 110 14400 3 b112 m111 u11 120 0 2 b111 m11 u12 111 3600 1 b11 m12 u111 112 7200 0 b12 m111 u112 111 10800 random sort $ hg log --rev 'sort(all(), "random")' | wc -l \s*8 (re) $ hg log --rev 'sort(all(), "-random")' | wc -l \s*8 (re) $ hg log --rev 'sort(all(), "random", random.seed=celeste)' 0 b12 m111 u112 111 10800 4 b111 m112 u111 110 14400 2 b111 m11 u12 111 3600 6 b111 t2 tu 130 0 1 b11 m12 u111 112 7200 7 b111 t3 tu 130 0 5 b111 t1 tu 130 0 3 b112 m111 u11 120 0 $ hg log --rev 'first(sort(all(), "random", random.seed=celeste))' 0 b12 m111 u112 111 10800 topographical sorting can't be combined with other sort keys, and you can't use the topo.firstbranch option when topo sort is not active: $ hg log -r 'sort(all(), "topo user")' hg: parse error: topo sort order cannot be combined with other sort keys [10] $ hg log -r 'sort(all(), user, topo.firstbranch=book1)' hg: parse error: topo.firstbranch can only be used when using the topo sort key [10] topo.firstbranch should accept any kind of expressions: $ hg log -r 'sort(0, topo, topo.firstbranch=(book1))' 0 b12 m111 u112 111 10800 $ cd .. $ cd repo test multiline revset with errors $ echo > multiline-revset $ echo '. +' >> multiline-revset $ echo '.^ +' >> multiline-revset $ hg log -r "`cat multiline-revset`" hg: parse error at 9: not a prefix: end ( . + .^ + ^ here) [10] $ hg debugrevspec -v 'revset(first(rev(0)))' -p all * parsed: (func (symbol 'revset') (func (symbol 'first') (func (symbol 'rev') (symbol '0')))) * expanded: (func (symbol 'revset') (func (symbol 'first') (func (symbol 'rev') (symbol '0')))) * concatenated: (func (symbol 'revset') (func (symbol 'first') (func (symbol 'rev') (symbol '0')))) * analyzed: (func (symbol 'revset') (func (symbol 'first') (func (symbol 'rev') (symbol '0')))) * optimized: (func (symbol 'revset') (func (symbol 'first') (func (symbol 'rev') (symbol '0')))) * set: 0 abort if the revset doesn't expect given size $ log 'expectsize()' hg: parse error: invalid set of arguments [10] $ log 'expectsize(0:2, a)' hg: parse error: expectsize requires a size range or a positive integer [10] $ log 'expectsize(0:2, 3)' 0 1 2 $ log 'expectsize(2:0, 3)' 2 1 0 $ log 'expectsize(0:1, 1)' abort: revset size mismatch. expected 1, got 2 [10] $ log 'expectsize(0:4, -1)' hg: parse error: negative size [10] $ log 'expectsize(0:2, 2:4)' 0 1 2 $ log 'expectsize(0:1, 3:5)' abort: revset size mismatch. expected between 3 and 5, got 2 [10] $ log 'expectsize(0:1, -1:2)' hg: parse error: negative size [10] $ log 'expectsize(0:1, 1:-2)' hg: parse error: negative size [10] $ log 'expectsize(0:2, a:4)' hg: parse error: size range bounds must be integers [10] $ log 'expectsize(0:2, 2:b)' hg: parse error: size range bounds must be integers [10] $ log 'expectsize(0:2, 2:)' 0 1 2 $ log 'expectsize(0:2, :5)' 0 1 2 $ log 'expectsize(0:2, :)' 0 1 2 $ log 'expectsize(0:2, 4:)' abort: revset size mismatch. expected between 4 and 11, got 3 [10] $ log 'expectsize(0:2, :2)' abort: revset size mismatch. expected between 0 and 2, got 3 [10] Test getting list of node from file $ hg log -r '0:2' -T '{node}\n' > some.nodes $ hg log -r 'nodefromfile("some.nodes")' -T '{rev}\n' 0 1 2 $ hg log -r 'nodefromfile("missing-file")' -T '{rev}\n' abort: cannot open nodes file "missing-file": $ENOENT$ [255] $ echo bad-node > bad.nodes $ hg log -r 'nodefromfile("bad.nodes")' -T '{rev}\n' $ echo abcdefabcdefabcdeabcdeabcdeabcdeabcdeabc > missing.nodes