Script for exporting WPT

third_party/WebKit/Tools/Scripts/webkitpy/w3c/sync.py

Index: third_party/WebKit/Tools/Scripts/webkitpy/w3c/sync.py
diff --git a/third_party/WebKit/Tools/Scripts/webkitpy/w3c/sync.py b/third_party/WebKit/Tools/Scripts/webkitpy/w3c/sync.py
new file mode 100644
index 0000000000000000000000000000000000000000..fd649f8b61c2c30f41031004db547bb139564e36
--- /dev/null
+++ b/third_party/WebKit/Tools/Scripts/webkitpy/w3c/sync.py
@@ -0,0 +1,145 @@
+# TODO: Do we need a license at the top of the file?

[foolip, 2016/10/25 19:31:41]

Yep, and the other one too.  Use the short one with "The Chromium Authors".

+
+from webkitpy.common.host import Host
+from webkitpy.common.webkit_finder import WebKitFinder
+import argparse
+import re
+import os
+
+
+WPT_REPO_URL = 'https://chromium.googlesource.com/external/w3c/web-platform-tests.git'
+WPT_DEST_NAME = 'wpt'
+CR_WPT_DIR = 'third_party/WebKit/LayoutTests/imported/wpt/'
+OLDEST_POSSIBLE_WPT_COMMIT = 'f2f9d854afdf0e9973c3bc4138ea41566bca2ef2'
+OLDEST_POSSIBLE_CR_COMMIT = '04ad58d09c1f6c46e888e42030b0566ac65411f6'  # before which no exporting should happen
+
+# Thoughts
+# - I feel like it would be cleaner to create the local WPT checkout outside of the local Chromium working tree. e.g. /tmp/wpt

[foolip, 2016/10/25 19:31:41]

Is this now doing the same thing as the importer, or what does
self.finder.path_from_webkit_base(WPT_DEST_NAME) do?

[jeffcarp, 2016/10/26 20:19:17]

It's currently putting the tests in third_party/WebKit/wpt (I'm not super
familiar with what the importer does)

+# - Maybe create a LocalWPT class? LocalWPT.clean(), LocalWPT.apply_patch()
+# - Idea: could use `with local_wpt(dir):` to manage setup/teardown
+
+
+class TestExporter(object):
+
+    def __init__(self, host, options):
+        self.host = host
+        self.options = options
+        self.fs = self.host.filesystem
+        self.finder = WebKitFinder(self.host.filesystem)
+        self.temp_repo_path = self.finder.path_from_webkit_base(WPT_DEST_NAME)
+
+    def run(self):
+
+        if not self.options['no_fetch']:

[foolip, 2016/10/25 19:31:41]

I can't see where this is hooked up, but I guess this and a lot is placeholders
still?

[jeffcarp, 2016/10/26 20:19:17]

Down on line 141 of this file there's an arg parser for it.

+            self.pull_down_local_wpt()
+
+        cr_export_commit = self.find_most_recent_wpt_export()
+
+        if cr_export_commit:
+            cr_commits = self.cr_commits_since(cr_export_commit)
+
+            def has_changes_in_wpt(commit):
+                assert commit
+
+                diff_files = self.host.executive.run_command([
+                    'git', 'diff-tree', '--no-commit-id',
+                    '--name-only', '-r', '{}'.format(commit)])
+
+                files = filter(bool, diff_files.split('\n'))
+                # Include files in LayoutTests/imported/wpt but not those with expectations
+                return any([CR_WPT_DIR in f and '-expected' not in f for f in files])

[foolip, 2016/10/25 19:31:41]

Maybe more strictly require the path to begin with CR_WPT_DIR

+
+            exportable_commits = filter(has_changes_in_wpt, cr_commits)
+
+            self.host.print_('## Found %d exportable commits in Chromium!' % len(exportable_commits))
+
+            for commit in exportable_commits:
+                # create da PR
+                # generate the patch
+                abs_cr_wpt_dir = os.path.abspath(os.path.join('..', '..', 'LayoutTests', 'imported', 'wpt'))
+                # Get patch but only for files in LayoutTests/imported/wpt
+                patch = self.host.executive.run_command([

[foolip, 2016/10/25 19:31:41]

Being able to get this patch on stdout or into a file will be useful at some
point probably, when this scripts fails to apply it. 

+                    'git', 'format-patch', '-1', '--stdout', '{}'.format(commit), abs_cr_wpt_dir])
+
+                self.run_command_in_wpt(['git', 'reset', '--hard', 'HEAD'])
+                self.run_command_in_wpt(['git', 'checkout', 'master'])
+                branch_name = 'chromium-try-{}'.format(commit)
+                remote_branch_name = 'remotes/github/%s' % branch_name
+
+                all_branches = map(str.strip, self.run_command_in_wpt(['git', 'branch', '-a']).split('\n'))
+
+                if branch_name in all_branches:
+                    self.host.print_('## Local branch %s already exists, skipping' % branch_name)
+                    continue
+                    # delete?
+                    # self.run_command_in_wpt(['git', 'branch', '-D', branch_name])
+
+                if remote_branch_name in all_branches:
+                    self.host.print_('## Remote branch %s already exists, skipping' % branch_name)
+                    continue
+
+                self.host.print_('## Creating local branch %s' % branch_name)
+                self.run_command_in_wpt(['git', 'checkout', '-b', branch_name])
+
+                # Remove Chromium WPT directory prefix
+                patch = patch.replace(CR_WPT_DIR, '')

[foolip, 2016/10/25 19:31:41]

Could also use git am -p<n> where n is len(CR_WPT_DIR.split(/')) or something
not off-by-one.

+                self.run_command_in_wpt(['git', 'apply', '-'], input=patch)
+
+                self.run_command_in_wpt(['git', 'commit', '-am', 'Testing export from Chromium'])
+                self.run_command_in_wpt(['git', 'push', 'github', branch_name])
+
+                # create PR on GH
+
+    def run_command_in_wpt(self, command, **kwargs):
+        return self.host.executive.run_command(command, cwd=self.temp_repo_path, **kwargs)
+
+    def pull_down_local_wpt(self):
+        url = WPT_REPO_URL
+
+        if self.fs.exists(self.temp_repo_path):
+            self.host.print_('## WPT checkout exists at %s, fetching latest' % (self.temp_repo_path))
+            self.run_command_in_wpt(['git', 'checkout', 'master'])
+            self.run_command_in_wpt(['git', 'fetch', '--all'])
+            self.run_command_in_wpt(['git', 'merge', '--ff-only', 'origin/master'])
+        else:
+            self.host.print_('## Cloning %s into %s' % (url, self.temp_repo_path))
+            self.host.executive.run_command(['git', 'clone', url, self.temp_repo_path])
+
+    def find_most_recent_wpt_export(self):
+        # The diff for the entire history of WPT is 3MB, which works fine for now

[foolip, 2016/10/25 19:31:41]

Works for now. Later, maybe git rev-list OLDEST_POSSIBLE_WPT_COMMIT..HEAD and
call git show -s --format=%B $commit for every commit until you find one that
matches.

[jeffcarp, 2016/10/26 22:06:12]

I had a possibly even simpler idea: why not just call:
git show -s --format=%B HEAD~0
git show -s --format=%B HEAD~1
git show -s --format=%B HEAD~2
git show -s --format=%B HEAD~3
.. and so on?

[foolip, 2016/10/26 22:34:36]

That'd require you to parse out the commit to terminate, but if we just export a
single whitespace change this script can just assume something will be found and
drop OLDEST_POSSIBLE_WPT_COMMIT.

One problem is that WPT's history is not linear, and the ~N syntax takes the
first parent of each commit, anything that came via a merge will be missed. I
think we normally won't merge, but it's probably just a question of time before
someone does it when resolving a conflict manually. It'd even be preferable for
non-trivial conflicts, to create a branch where it applies cleanly, and then
resolve conflicts in the merge commit.

Merges might also require some finesse in determining what the latest commit is.
I don't think --topo-order will do it, but maybe --author-date-order if we're
not doing anything weird.

+        # (Picked a random commit)
+        diff_output = self.host.executive.run_command([
+            'git', 'log', '1d81fe1feb8f9dc43bc101a4cf9d755fdf4df0e7..HEAD'

[foolip, 2016/10/25 19:31:41]

Could use OLDEST_POSSIBLE_WPT_COMMIT here too?

+        ], cwd=self.temp_repo_path)
+
+        commit_messages = diff_output.split('\ncommit ')
+
+        for message in commit_messages:
+            commit = message.split('\n')[0].replace('commit ', '')
+            if commit == OLDEST_POSSIBLE_WPT_COMMIT:
+                self.host.print_('## Reached oldest possible WPT before finding a Chromium export, stopping')
+                return None
+            elif 'Cr-Commit-Position' in message:
+                match = re.match(r'Cr-Commit-Position: (?P<cr_commit>.+)$', message)
+                cr_commit = match.group('cr_commit')
+                if cr_commit:
+                    self.host.print_('## Found last exported commit: web-platform-tests@%s crrev: %s' % commit, cr_commit)
+                    return cr_commit
+            elif 'testharness' in message:
+                # Fake case for testing
+                return OLDEST_POSSIBLE_CR_COMMIT
+
+    def cr_commits_since(self, cr_commit):
+        commits = self.host.executive.run_command([
+            'git', 'rev-list', '{}..HEAD'.format(cr_commit)])

[foolip, 2016/10/25 19:31:41]

Will this not need --reverse to test them in the right order?

[jeffcarp, 2016/10/26 20:19:17]

Ah thanks for catching that.

+        return filter(bool, commits.split('\n'))
+
+
+def main():
+    parser = argparse.ArgumentParser(description='WPT bridge bot')
+
+    parser.add_argument('--no-fetch', action='store_true')
+
+    options = parser.parse_args()
+    test_exporter = TestExporter(host=Host(), options=options)
+    test_exporter.run()