bundle2: store changeset count when creating file bundles
The bundle2 changegroup part has an advisory param saying how many
changesets are in the part. Before this patch, we were setting
this part when generating bundle2 parts via the wire protocol but
not when generating local bundle2 files.
A side effect of not setting the changeset count part is that progress
bars don't work when applying changesets. As the tests show, this
impacted clone bundles, shelve, backup bundles, `hg unbundle`, and
anything touching bundle2 files.
This patch adds a backdoor to allow us to pass state from
changegroup generation into the unbundler. We store the number
of changesets in the changegroup in this state and use it to
populate the aforementioned advisory part parameter when generating
the bundle2 bundle.
I concede that I'm not thrilled by how state is being passed in
changegroup.py (it feels a bit hacky). I would love to overhaul the
rather confusing set of functions in changegroup.py with something that
passes rich objects around instead of e.g. low-level generators.
However, given the code freeze for 3.9 is imminent, I'd rather not
undertake this endeavor right now. This feels like the easiest way
to get the parameter added to the changegroup part.
This file tests the behavior of run-tests.py itself.
Avoid interference from actual test env:
$ . "$TESTDIR/helper-runtests.sh"
Smoke test with install
============
$ run-tests.py $HGTEST_RUN_TESTS_PURE -l
# Ran 0 tests, 0 skipped, 0 warned, 0 failed.
Define a helper to avoid the install step
=============
$ rt()
> {
> run-tests.py --with-hg=`which hg` "$@"
> }
error paths
#if symlink
$ ln -s `which true` hg
$ run-tests.py --with-hg=./hg
warning: --with-hg should specify an hg script
# Ran 0 tests, 0 skipped, 0 warned, 0 failed.
$ rm hg
#endif
#if execbit
$ touch hg
$ run-tests.py --with-hg=./hg
Usage: run-tests.py [options] [tests]
run-tests.py: error: --with-hg must specify an executable hg script
[2]
$ rm hg
#endif
an empty test
=======================
$ touch test-empty.t
$ rt
.
# Ran 1 tests, 0 skipped, 0 warned, 0 failed.
$ rm test-empty.t
a succesful test
=======================
$ cat > test-success.t << EOF
> $ echo babar
> babar
> $ echo xyzzy
> never*happens (glob) (?)
> xyzzy
> nor this (?)
> $ printf 'abc\ndef\nxyz\n'
> 123 (?)
> abc
> def (?)
> 456 (?)
> xyz
> EOF
$ rt
.
# Ran 1 tests, 0 skipped, 0 warned, 0 failed.
failing test
==================
test churn with globs
$ cat > test-failure.t <<EOF
> $ echo "bar-baz"; echo "bar-bad"
> bar*bad (glob)
> bar*baz (glob)
> EOF
$ rt test-failure.t
--- $TESTTMP/test-failure.t
+++ $TESTTMP/test-failure.t.err
@@ -1,3 +1,3 @@
$ echo "bar-baz"; echo "bar-bad"
+ bar*baz (glob)
bar*bad (glob)
- bar*baz (glob)
ERROR: test-failure.t output changed
!
Failed test-failure.t: output changed
# Ran 1 tests, 0 skipped, 0 warned, 1 failed.
python hash seed: * (glob)
[1]
basic failing test
$ cat > test-failure.t << EOF
> $ echo babar
> rataxes
> This is a noop statement so that
> this test is still more bytes than success.
> pad pad pad pad............................................................
> pad pad pad pad............................................................
> EOF
>>> fh = open('test-failure-unicode.t', 'wb')
>>> fh.write(u' $ echo babar\u03b1\n'.encode('utf-8')) and None
>>> fh.write(u' l\u03b5\u03b5t\n'.encode('utf-8')) and None
$ rt
--- $TESTTMP/test-failure.t
+++ $TESTTMP/test-failure.t.err
@@ -1,5 +1,5 @@
$ echo babar
- rataxes
+ babar
This is a noop statement so that
this test is still more bytes than success.
pad pad pad pad............................................................
ERROR: test-failure.t output changed
!.
--- $TESTTMP/test-failure-unicode.t
+++ $TESTTMP/test-failure-unicode.t.err
@@ -1,2 +1,2 @@
$ echo babar\xce\xb1 (esc)
- l\xce\xb5\xce\xb5t (esc)
+ babar\xce\xb1 (esc)
ERROR: test-failure-unicode.t output changed
!
Failed test-failure.t: output changed
Failed test-failure-unicode.t: output changed
# Ran 3 tests, 0 skipped, 0 warned, 2 failed.
python hash seed: * (glob)
[1]
test --xunit support
$ rt --xunit=xunit.xml
--- $TESTTMP/test-failure.t
+++ $TESTTMP/test-failure.t.err
@@ -1,5 +1,5 @@
$ echo babar
- rataxes
+ babar
This is a noop statement so that
this test is still more bytes than success.
pad pad pad pad............................................................
ERROR: test-failure.t output changed
!.
--- $TESTTMP/test-failure-unicode.t
+++ $TESTTMP/test-failure-unicode.t.err
@@ -1,2 +1,2 @@
$ echo babar\xce\xb1 (esc)
- l\xce\xb5\xce\xb5t (esc)
+ babar\xce\xb1 (esc)
ERROR: test-failure-unicode.t output changed
!
Failed test-failure.t: output changed
Failed test-failure-unicode.t: output changed
# Ran 3 tests, 0 skipped, 0 warned, 2 failed.
python hash seed: * (glob)
[1]
$ cat xunit.xml
<?xml version="1.0" encoding="utf-8"?>
<testsuite errors="0" failures="2" name="run-tests" skipped="0" tests="3">
<testcase name="test-success.t" time="*"/> (glob)
<testcase name="test-failure-unicode.t" time="*"> (glob)
<![CDATA[--- $TESTTMP/test-failure-unicode.t
+++ $TESTTMP/test-failure-unicode.t.err
@@ -1,2 +1,2 @@
$ echo babar\xce\xb1 (esc)
- l\xce\xb5\xce\xb5t (esc)
+ babar\xce\xb1 (esc)
]]> </testcase>
<testcase name="test-failure.t" time="*"> (glob)
<![CDATA[--- $TESTTMP/test-failure.t
+++ $TESTTMP/test-failure.t.err
@@ -1,5 +1,5 @@
$ echo babar
- rataxes
+ babar
This is a noop statement so that
this test is still more bytes than success.
pad pad pad pad............................................................
]]> </testcase>
</testsuite>
$ cat .testtimes
test-failure-unicode.t * (glob)
test-failure.t * (glob)
test-success.t * (glob)
$ rm test-failure-unicode.t
test for --retest
====================
$ rt --retest
--- $TESTTMP/test-failure.t
+++ $TESTTMP/test-failure.t.err
@@ -1,5 +1,5 @@
$ echo babar
- rataxes
+ babar
This is a noop statement so that
this test is still more bytes than success.
pad pad pad pad............................................................
ERROR: test-failure.t output changed
!
Failed test-failure.t: output changed
# Ran 2 tests, 1 skipped, 0 warned, 1 failed.
python hash seed: * (glob)
[1]
Selecting Tests To Run
======================
successful
$ rt test-success.t
.
# Ran 1 tests, 0 skipped, 0 warned, 0 failed.
success w/ keyword
$ rt -k xyzzy
.
# Ran 2 tests, 1 skipped, 0 warned, 0 failed.
failed
$ rt test-failure.t
--- $TESTTMP/test-failure.t
+++ $TESTTMP/test-failure.t.err
@@ -1,5 +1,5 @@
$ echo babar
- rataxes
+ babar
This is a noop statement so that
this test is still more bytes than success.
pad pad pad pad............................................................
ERROR: test-failure.t output changed
!
Failed test-failure.t: output changed
# Ran 1 tests, 0 skipped, 0 warned, 1 failed.
python hash seed: * (glob)
[1]
failure w/ keyword
$ rt -k rataxes
--- $TESTTMP/test-failure.t
+++ $TESTTMP/test-failure.t.err
@@ -1,5 +1,5 @@
$ echo babar
- rataxes
+ babar
This is a noop statement so that
this test is still more bytes than success.
pad pad pad pad............................................................
ERROR: test-failure.t output changed
!
Failed test-failure.t: output changed
# Ran 2 tests, 1 skipped, 0 warned, 1 failed.
python hash seed: * (glob)
[1]
Verify that when a process fails to start we show a useful message
==================================================================
$ cat > test-serve-fail.t <<EOF
> $ echo 'abort: child process failed to start blah'
> EOF
$ rt test-serve-fail.t
ERROR: test-serve-fail.t output changed
!
Failed test-serve-fail.t: server failed to start (HGPORT=*) (glob)
# Ran 1 tests, 0 skipped, 0 warned, 1 failed.
python hash seed: * (glob)
[1]
$ rm test-serve-fail.t
Verify that we can try other ports
===================================
$ hg init inuse
$ hg serve -R inuse -p $HGPORT -d --pid-file=blocks.pid
$ cat blocks.pid >> $DAEMON_PIDS
$ cat > test-serve-inuse.t <<EOF
> $ hg serve -R `pwd`/inuse -p \$HGPORT -d --pid-file=hg.pid
> $ cat hg.pid >> \$DAEMON_PIDS
> EOF
$ rt test-serve-inuse.t
.
# Ran 1 tests, 0 skipped, 0 warned, 0 failed.
$ rm test-serve-inuse.t
$ killdaemons.py $DAEMON_PIDS
$ rm $DAEMON_PIDS
Running In Debug Mode
======================
$ rt --debug 2>&1 | grep -v pwd
+ echo *SALT* 0 0 (glob)
*SALT* 0 0 (glob)
+ echo babar
babar
+ echo *SALT* 6 0 (glob)
*SALT* 6 0 (glob)
*+ echo *SALT* 0 0 (glob)
*SALT* 0 0 (glob)
+ echo babar
babar
+ echo *SALT* 2 0 (glob)
*SALT* 2 0 (glob)
+ echo xyzzy
xyzzy
+ echo *SALT* 6 0 (glob)
*SALT* 6 0 (glob)
+ printf *abc\ndef\nxyz\n* (glob)
abc
def
xyz
+ echo *SALT* 12 0 (glob)
*SALT* 12 0 (glob)
.
# Ran 2 tests, 0 skipped, 0 warned, 0 failed.
Parallel runs
==============
(duplicate the failing test to get predictable output)
$ cp test-failure.t test-failure-copy.t
$ rt --jobs 2 test-failure*.t -n
!!
Failed test-failure*.t: output changed (glob)
Failed test-failure*.t: output changed (glob)
# Ran 2 tests, 0 skipped, 0 warned, 2 failed.
python hash seed: * (glob)
[1]
failures in parallel with --first should only print one failure
>>> f = open('test-nothing.t', 'w')
>>> f.write('foo\n' * 1024) and None
>>> f.write(' $ sleep 1') and None
$ rt --jobs 2 --first
--- $TESTTMP/test-failure*.t (glob)
+++ $TESTTMP/test-failure*.t.err (glob)
@@ -1,5 +1,5 @@
$ echo babar
- rataxes
+ babar
This is a noop statement so that
this test is still more bytes than success.
pad pad pad pad............................................................
Failed test-failure*.t: output changed (glob)
Failed test-nothing.t: output changed
# Ran 2 tests, 0 skipped, 0 warned, 2 failed.
python hash seed: * (glob)
[1]
(delete the duplicated test file)
$ rm test-failure-copy.t test-nothing.t
Interactive run
===============
(backup the failing test)
$ cp test-failure.t backup
Refuse the fix
$ echo 'n' | rt -i
--- $TESTTMP/test-failure.t
+++ $TESTTMP/test-failure.t.err
@@ -1,5 +1,5 @@
$ echo babar
- rataxes
+ babar
This is a noop statement so that
this test is still more bytes than success.
pad pad pad pad............................................................
Accept this change? [n]
ERROR: test-failure.t output changed
!.
Failed test-failure.t: output changed
# Ran 2 tests, 0 skipped, 0 warned, 1 failed.
python hash seed: * (glob)
[1]
$ cat test-failure.t
$ echo babar
rataxes
This is a noop statement so that
this test is still more bytes than success.
pad pad pad pad............................................................
pad pad pad pad............................................................
Interactive with custom view
$ echo 'n' | rt -i --view echo
$TESTTMP/test-failure.t $TESTTMP/test-failure.t.err (glob)
Accept this change? [n]* (glob)
ERROR: test-failure.t output changed
!.
Failed test-failure.t: output changed
# Ran 2 tests, 0 skipped, 0 warned, 1 failed.
python hash seed: * (glob)
[1]
View the fix
$ echo 'y' | rt --view echo
$TESTTMP/test-failure.t $TESTTMP/test-failure.t.err (glob)
ERROR: test-failure.t output changed
!.
Failed test-failure.t: output changed
# Ran 2 tests, 0 skipped, 0 warned, 1 failed.
python hash seed: * (glob)
[1]
Accept the fix
$ echo " $ echo 'saved backup bundle to \$TESTTMP/foo.hg'" >> test-failure.t
$ echo " saved backup bundle to \$TESTTMP/foo.hg" >> test-failure.t
$ echo " $ echo 'saved backup bundle to \$TESTTMP/foo.hg'" >> test-failure.t
$ echo " saved backup bundle to \$TESTTMP/foo.hg (glob)" >> test-failure.t
$ echo " $ echo 'saved backup bundle to \$TESTTMP/foo.hg'" >> test-failure.t
$ echo " saved backup bundle to \$TESTTMP/*.hg (glob)" >> test-failure.t
$ echo 'y' | rt -i 2>&1
--- $TESTTMP/test-failure.t
+++ $TESTTMP/test-failure.t.err
@@ -1,11 +1,11 @@
$ echo babar
- rataxes
+ babar
This is a noop statement so that
this test is still more bytes than success.
pad pad pad pad............................................................
pad pad pad pad............................................................
$ echo 'saved backup bundle to $TESTTMP/foo.hg'
- saved backup bundle to $TESTTMP/foo.hg
+ saved backup bundle to $TESTTMP/foo.hg* (glob)
$ echo 'saved backup bundle to $TESTTMP/foo.hg'
saved backup bundle to $TESTTMP/foo.hg* (glob)
$ echo 'saved backup bundle to $TESTTMP/foo.hg'
Accept this change? [n] ..
# Ran 2 tests, 0 skipped, 0 warned, 0 failed.
$ sed -e 's,(glob)$,&<,g' test-failure.t
$ echo babar
babar
This is a noop statement so that
this test is still more bytes than success.
pad pad pad pad............................................................
pad pad pad pad............................................................
$ echo 'saved backup bundle to $TESTTMP/foo.hg'
saved backup bundle to $TESTTMP/foo.hg (glob)<
$ echo 'saved backup bundle to $TESTTMP/foo.hg'
saved backup bundle to $TESTTMP/foo.hg (glob)<
$ echo 'saved backup bundle to $TESTTMP/foo.hg'
saved backup bundle to $TESTTMP/*.hg (glob)<
(reinstall)
$ mv backup test-failure.t
No Diff
===============
$ rt --nodiff
!.
Failed test-failure.t: output changed
# Ran 2 tests, 0 skipped, 0 warned, 1 failed.
python hash seed: * (glob)
[1]
test --tmpdir support
$ rt --tmpdir=$TESTTMP/keep test-success.t
Keeping testtmp dir: $TESTTMP/keep/child1/test-success.t (glob)
Keeping threadtmp dir: $TESTTMP/keep/child1 (glob)
.
# Ran 1 tests, 0 skipped, 0 warned, 0 failed.
timeouts
========
$ cat > test-timeout.t <<EOF
> $ sleep 2
> $ echo pass
> pass
> EOF
> echo '#require slow' > test-slow-timeout.t
> cat test-timeout.t >> test-slow-timeout.t
$ rt --timeout=1 --slowtimeout=3 test-timeout.t test-slow-timeout.t
st
Skipped test-slow-timeout.t: missing feature: allow slow tests
Failed test-timeout.t: timed out
# Ran 1 tests, 1 skipped, 0 warned, 1 failed.
python hash seed: * (glob)
[1]
$ rt --timeout=1 --slowtimeout=3 \
> test-timeout.t test-slow-timeout.t --allow-slow-tests
.t
Failed test-timeout.t: timed out
# Ran 2 tests, 0 skipped, 0 warned, 1 failed.
python hash seed: * (glob)
[1]
$ rm test-timeout.t test-slow-timeout.t
test for --time
==================
$ rt test-success.t --time
.
# Ran 1 tests, 0 skipped, 0 warned, 0 failed.
# Producing time report
start end cuser csys real Test
\s*[\d\.]{5} \s*[\d\.]{5} \s*[\d\.]{5} \s*[\d\.]{5} \s*[\d\.]{5} test-success.t (re)
test for --time with --job enabled
====================================
$ rt test-success.t --time --jobs 2
.
# Ran 1 tests, 0 skipped, 0 warned, 0 failed.
# Producing time report
start end cuser csys real Test
\s*[\d\.]{5} \s*[\d\.]{5} \s*[\d\.]{5} \s*[\d\.]{5} \s*[\d\.]{5} test-success.t (re)
Skips
================
$ cat > test-skip.t <<EOF
> $ echo xyzzy
> #require false
> EOF
$ rt --nodiff
!.s
Skipped test-skip.t: missing feature: nail clipper
Failed test-failure.t: output changed
# Ran 2 tests, 1 skipped, 0 warned, 1 failed.
python hash seed: * (glob)
[1]
$ rt --keyword xyzzy
.s
Skipped test-skip.t: missing feature: nail clipper
# Ran 2 tests, 2 skipped, 0 warned, 0 failed.
Skips with xml
$ rt --keyword xyzzy \
> --xunit=xunit.xml
.s
Skipped test-skip.t: missing feature: nail clipper
# Ran 2 tests, 2 skipped, 0 warned, 0 failed.
$ cat xunit.xml
<?xml version="1.0" encoding="utf-8"?>
<testsuite errors="0" failures="0" name="run-tests" skipped="2" tests="2">
<testcase name="test-success.t" time="*"/> (glob)
</testsuite>
Missing skips or blacklisted skips don't count as executed:
$ echo test-failure.t > blacklist
$ rt --blacklist=blacklist --json\
> test-failure.t test-bogus.t
ss
Skipped test-bogus.t: Doesn't exist
Skipped test-failure.t: blacklisted
# Ran 0 tests, 2 skipped, 0 warned, 0 failed.
$ cat report.json
testreport ={
"test-bogus.t": {
"result": "skip"
},
"test-failure.t": {
"result": "skip"
}
} (no-eol)
Whitelist trumps blacklist
$ echo test-failure.t > whitelist
$ rt --blacklist=blacklist --whitelist=whitelist --json\
> test-failure.t test-bogus.t
s
--- $TESTTMP/test-failure.t
+++ $TESTTMP/test-failure.t.err
@@ -1,5 +1,5 @@
$ echo babar
- rataxes
+ babar
This is a noop statement so that
this test is still more bytes than success.
pad pad pad pad............................................................
ERROR: test-failure.t output changed
!
Skipped test-bogus.t: Doesn't exist
Failed test-failure.t: output changed
# Ran 1 tests, 1 skipped, 0 warned, 1 failed.
python hash seed: * (glob)
[1]
test for --json
==================
$ rt --json
--- $TESTTMP/test-failure.t
+++ $TESTTMP/test-failure.t.err
@@ -1,5 +1,5 @@
$ echo babar
- rataxes
+ babar
This is a noop statement so that
this test is still more bytes than success.
pad pad pad pad............................................................
ERROR: test-failure.t output changed
!.s
Skipped test-skip.t: missing feature: nail clipper
Failed test-failure.t: output changed
# Ran 2 tests, 1 skipped, 0 warned, 1 failed.
python hash seed: * (glob)
[1]
$ cat report.json
testreport ={
"test-failure.t": [\{] (re)
"csys": "\s*[\d\.]{4,5}", ? (re)
"cuser": "\s*[\d\.]{4,5}", ? (re)
"diff": "---.+\+\+\+.+", ? (re)
"end": "\s*[\d\.]{4,5}", ? (re)
"result": "failure", ? (re)
"start": "\s*[\d\.]{4,5}", ? (re)
"time": "\s*[\d\.]{4,5}" (re)
}, ? (re)
"test-skip.t": {
"csys": "\s*[\d\.]{4,5}", ? (re)
"cuser": "\s*[\d\.]{4,5}", ? (re)
"diff": "", ? (re)
"end": "\s*[\d\.]{4,5}", ? (re)
"result": "skip", ? (re)
"start": "\s*[\d\.]{4,5}", ? (re)
"time": "\s*[\d\.]{4,5}" (re)
}, ? (re)
"test-success.t": [\{] (re)
"csys": "\s*[\d\.]{4,5}", ? (re)
"cuser": "\s*[\d\.]{4,5}", ? (re)
"diff": "", ? (re)
"end": "\s*[\d\.]{4,5}", ? (re)
"result": "success", ? (re)
"start": "\s*[\d\.]{4,5}", ? (re)
"time": "\s*[\d\.]{4,5}" (re)
}
} (no-eol)
Test that failed test accepted through interactive are properly reported:
$ cp test-failure.t backup
$ echo y | rt --json -i
--- $TESTTMP/test-failure.t
+++ $TESTTMP/test-failure.t.err
@@ -1,5 +1,5 @@
$ echo babar
- rataxes
+ babar
This is a noop statement so that
this test is still more bytes than success.
pad pad pad pad............................................................
Accept this change? [n] ..s
Skipped test-skip.t: missing feature: nail clipper
# Ran 2 tests, 1 skipped, 0 warned, 0 failed.
$ cat report.json
testreport ={
"test-failure.t": [\{] (re)
"csys": "\s*[\d\.]{4,5}", ? (re)
"cuser": "\s*[\d\.]{4,5}", ? (re)
"diff": "", ? (re)
"end": "\s*[\d\.]{4,5}", ? (re)
"result": "success", ? (re)
"start": "\s*[\d\.]{4,5}", ? (re)
"time": "\s*[\d\.]{4,5}" (re)
}, ? (re)
"test-skip.t": {
"csys": "\s*[\d\.]{4,5}", ? (re)
"cuser": "\s*[\d\.]{4,5}", ? (re)
"diff": "", ? (re)
"end": "\s*[\d\.]{4,5}", ? (re)
"result": "skip", ? (re)
"start": "\s*[\d\.]{4,5}", ? (re)
"time": "\s*[\d\.]{4,5}" (re)
}, ? (re)
"test-success.t": [\{] (re)
"csys": "\s*[\d\.]{4,5}", ? (re)
"cuser": "\s*[\d\.]{4,5}", ? (re)
"diff": "", ? (re)
"end": "\s*[\d\.]{4,5}", ? (re)
"result": "success", ? (re)
"start": "\s*[\d\.]{4,5}", ? (re)
"time": "\s*[\d\.]{4,5}" (re)
}
} (no-eol)
$ mv backup test-failure.t
backslash on end of line with glob matching is handled properly
$ cat > test-glob-backslash.t << EOF
> $ echo 'foo bar \\'
> foo * \ (glob)
> EOF
$ rt test-glob-backslash.t
.
# Ran 1 tests, 0 skipped, 0 warned, 0 failed.
$ rm -f test-glob-backslash.t
Test globbing of 127.0.0.1
$ echo 172.16.18.1
127.0.0.1 (glob)
Test reusability for third party tools
======================================
$ mkdir "$TESTTMP"/anothertests
$ cd "$TESTTMP"/anothertests
test that `run-tests.py` can execute hghave, even if it runs not in
Mercurial source tree.
$ cat > test-hghave.t <<EOF
> #require true
> $ echo foo
> foo
> EOF
$ rt test-hghave.t
.
# Ran 1 tests, 0 skipped, 0 warned, 0 failed.
test that RUNTESTDIR refers the directory, in which `run-tests.py` now
running is placed.
$ cat > test-runtestdir.t <<EOF
> - $TESTDIR, in which test-run-tests.t is placed
> - \$TESTDIR, in which test-runtestdir.t is placed (expanded at runtime)
> - \$RUNTESTDIR, in which run-tests.py is placed (expanded at runtime)
>
> #if windows
> $ test "\$TESTDIR" = "$TESTTMP\anothertests"
> #else
> $ test "\$TESTDIR" = "$TESTTMP"/anothertests
> #endif
> $ test "\$RUNTESTDIR" = "$TESTDIR"
> $ head -n 3 "\$RUNTESTDIR"/../contrib/check-code.py
> #!/usr/bin/env python
> #
> # check-code - a style and portability checker for Mercurial
> EOF
$ rt test-runtestdir.t
.
# Ran 1 tests, 0 skipped, 0 warned, 0 failed.
#if execbit
test that TESTDIR is referred in PATH
$ cat > custom-command.sh <<EOF
> #!/bin/sh
> echo "hello world"
> EOF
$ chmod +x custom-command.sh
$ cat > test-testdir-path.t <<EOF
> $ custom-command.sh
> hello world
> EOF
$ rt test-testdir-path.t
.
# Ran 1 tests, 0 skipped, 0 warned, 0 failed.
#endif
test support for --allow-slow-tests
$ cat > test-very-slow-test.t <<EOF
> #require slow
> $ echo pass
> pass
> EOF
$ rt test-very-slow-test.t
s
Skipped test-very-slow-test.t: missing feature: allow slow tests
# Ran 0 tests, 1 skipped, 0 warned, 0 failed.
$ rt $HGTEST_RUN_TESTS_PURE --allow-slow-tests test-very-slow-test.t
.
# Ran 1 tests, 0 skipped, 0 warned, 0 failed.
support for running a test outside the current directory
$ mkdir nonlocal
$ cat > nonlocal/test-is-not-here.t << EOF
> $ echo pass
> pass
> EOF
$ rt nonlocal/test-is-not-here.t
.
# Ran 1 tests, 0 skipped, 0 warned, 0 failed.
support for bisecting failed tests automatically
$ hg init bisect
$ cd bisect
$ cat >> test-bisect.t <<EOF
> $ echo pass
> pass
> EOF
$ hg add test-bisect.t
$ hg ci -m 'good'
$ cat >> test-bisect.t <<EOF
> $ echo pass
> fail
> EOF
$ hg ci -m 'bad'
$ rt --known-good-rev=0 test-bisect.t
--- $TESTTMP/anothertests/bisect/test-bisect.t
+++ $TESTTMP/anothertests/bisect/test-bisect.t.err
@@ -1,4 +1,4 @@
$ echo pass
pass
$ echo pass
- fail
+ pass
ERROR: test-bisect.t output changed
!
Failed test-bisect.t: output changed
test-bisect.t broken by 72cbf122d116 (bad)
# Ran 1 tests, 0 skipped, 0 warned, 1 failed.
python hash seed: * (glob)
[1]