Commit parsers to handle merge commit prefixes

[TGNThump]

When you merge a PR within Azure DevOps, a prefix (Merged PR 12345:) is added to the commit message unless you manually remove it for every PR, leading to commit messages like:

Merged PR 541234: chore: Improve linting

Is there any way to specify a prefix regex to ignore when parsing commits or any other way to handle this case without requiring contributors to dig into the advanced merge options for every PR?

See conventional-changelog/standard-version#690 for a similar discussion from another project and https://developercommunity.visualstudio.com/t/change-default-title-for-pull-request-commits-to-n-1/365716

[bernardcooke53]

Hi @TGNThump,

Thanks for bringing this up - I don't think this is currently possible, but I think it would be a nice feature to have - a short term workaround would be to write a custom commit parser, but from the issue you referenced above it seems like this is something that multiple VCS do so it would be nice to have that option.

Adding an option to the commit parsers for an "optional merge commit pattern" which can be stripped should manage this.

[manudawber]

For anyone facing this with GitHub merge commits:

from semantic_release.history import angular_parser
from semantic_release.history.parser_helpers import ParsedCommit

MERGE_COMMIT_PREFIX = "Merge pull request"


def parse_commit_message(message: str) -> ParsedCommit:
    """Extends the `angular_parser` function to work with merge commits.

    Merge commits are not processed by `angular_parser` as they don't adhere to the conventional commit format. Given
    that we enforce pull request titles to follow this format, this custom parser is required to ensure these PR titles
    are usable by `python-semantic-release`.

    If a commit message starts with `MERGE_COMMIT_PREFIX` then it's considered a merge commit and the last line of the
    commit message will be extracted (this should be the original PR title) and sent to `angular_parser`.

    Example merge commit message:

    ```
    Merge pull request #553 from octocat/feature-x

    feat: enable feature X
    ```

    Args:
        message: A git commit message.

    Returns:
        A `ParsedCommit` object containing semantic release version and changelog information.
    """
    message = message.strip()  # remove leading/trailing blank lines
    if message.startswith(MERGE_COMMIT_PREFIX):
        message = message.splitlines()[-1]
    message = message.strip()  # remove leading/trailing whitespace
    return angular_parser(message)

[github-actions]

It has been 60 days since the last update on this confirmed issue. @python-semantic-release/team can you provide an update on the status of this issue?

[codejedi365]

No update yet, still in the backlog

[lawrenceeldridge]

Not sure if this helps but ended up creating a custom parser to work with Azure Devops merge commits. Hope this is useful until a dedicated option is available for prefix handling... just note that in this case I am using Poetry with pyproject.toml, so the tag configuration is being retrieved from here.

import re
import toml
import git

from semantic_release import (
    CommitParser,
    ParseResult,
    ParserOptions,
    ParsedCommit,
    LevelBump,
    ParseError,
)
from typing import Optional, Any


class MyParserOptions(ParserOptions):
    def __init__(self, **_: Any):
        # Load whole configuration
        super().__init__(**_)
        pyproject = toml.load("pyproject.toml")

        # Extract commit parsers options
        commit_parser_options = pyproject["tool"]["semantic_release"][
            "commit_parser_options"
        ]

        # Load options as instance variables
        self.allowed_tags = commit_parser_options["allowed_tags"]
        self.minor_tags = commit_parser_options["minor_tags"]
        self.patch_tags = commit_parser_options["patch_tags"]
        self.default_bump_level = LevelBump(commit_parser_options["default_bump_level"])


class MyCommitParser(CommitParser[ParseResult, MyParserOptions]):

    def __init__(self, **kwargs):
        super().__init__(**kwargs)
        self.options = MyParserOptions()

    def parse(self, commit: git.objects.commit.Commit) -> Optional[ParseResult]:
        sentences = commit.message.strip().split("\n\n")
        unique_sentences = list(dict.fromkeys(sentences))
        message = "\n\n".join(unique_sentences)
        allowed_tags_regex_part = "|".join(map(re.escape, self.options.allowed_tags))
        match = re.match(
            rf"^.*({allowed_tags_regex_part})(?:\(([^)]+)\))?: (.+)(?:\n|$)([\s\S]*)",
            message,
        )

        if match:
            type_ = match.group(1)
            scope = match.group(2) or ""
            description = match.group(3)
            rest_of_message = match.group(4)

            descriptions = [description]
            breaking_descriptions = []

            for line in rest_of_message.split("\n"):
                line = line.strip()
                if line.startswith("BREAKING CHANGE: "):
                    breaking_descriptions.append(line.replace("BREAKING CHANGE: ", ""))
                elif line != "":
                    descriptions.append(line)

            bump = self.options.default_bump_level
            if breaking_descriptions:
                bump = LevelBump.MAJOR
            elif type_ in self.options.minor_tags:
                bump = LevelBump.MINOR
            elif type_ in self.options.patch_tags:
                bump = LevelBump.PATCH

            return ParsedCommit(
                bump=bump,
                type=type_,
                scope=scope,
                descriptions=descriptions,
                breaking_descriptions=breaking_descriptions,
                commit=commit,
            )

        else:
            return ParseError(commit=commit, error="Could not parse commit message")

[github-actions]

It has been 60 days since the last update on this confirmed issue. @python-semantic-release/team can you provide an update on the status of this issue?

[mikenerone]

A generalized approach for having different options for what to do with merge commits would be helpful. At my org we annotate each commit fully, so the messages in the merge commits are actually redundant and I'd like to simply skip them when generating the CHANGELOG.

[codejedi365]

@mikenerone, I agree. You can see an example in PSR's project configuration. Merge commits are completely ignored using the changelog.exclude_commit_patterns.

[codejedi365]

Unless there is a compelling reason, I move to close this feature request because ultimately your merge commits are nothing but duplicates of commits already in the git tree. If you are using semantic commits as they are intended then the changes are already documented and merge commits should just be excluded using the pattern matches. I may in the future include a simple Boolean to identify and filter out merge commits but until then the pattern exclusion is the recommended method for handling merge commits in history.