Cap error fields in every error event and build release notes from CHANGELOG.md

CHANGELOG.md

@@ -67,6 +67,12 @@ under Unreleased and move into a version section at release time.
 
 ### Fixed
 
+- Every error event now caps `error_message` at 500 characters, keeping the
+  start and the end of a longer message. Integrations used to cut it to the
+  first 200, which lost the cause in messages that end with it, such as
+  mlx-whisper audio errors that start with the ffmpeg banner. Manual
+  `track_error()` calls had no limit at all. `provider_error_code` is capped
+  at 64 characters for every event.
 - The llama.cpp integration now tracks `Llama.create_chat_completion()` and
   `Llama.create_completion()`, streaming or not. It used to patch only
   `Llama.__call__`, so chat apps on llama-cpp-python reported model loads but

CONTRIBUTING.md

@@ -49,7 +49,7 @@
 Releases follow a `release/<version>` branching flow:
 
 1. Cut a `release/<version>` branch from `devel`.
-2. Bump the version in `pyproject.toml`.
+2. Bump the version in `pyproject.toml` and move the Unreleased entries in `CHANGELOG.md` into a `## <version> - <date>` section. The release notes are that section, so the release fails without it.
 3. Open a pull request targeting `main`. CI will automatically run the full compatibility matrix and post a changelog preview comment.
 4. Once merged, tag the commit as `v<version>`. The release workflow publishes to PyPI and creates a GitHub release.
 

scripts/build_changelog_comment.py

@@ -1,6 +1,9 @@
 #!/usr/bin/env python3
 """Generate changelog text for release PRs and for tag-build release notes.
 
+The notes are the version's section of CHANGELOG.md, so they list what changed
+for users rather than every commit and merge since the previous tag.
+
 The two callers differ only in wording: release-pr.yml renders a preview of
 notes that do not exist yet, while release.yml (which sets TAG_NAME) renders
 the notes themselves.
@@ -13,6 +16,8 @@
 import subprocess
 from pathlib import Path
 
+CHANGELOG = Path("CHANGELOG.md")
+
 
 def get_version() -> str:
     # Regex instead of tomllib so the script also runs on Python 3.10.
@@ -24,9 +29,6 @@ def get_version() -> str:
     return match.group(1)
 
 
-RELEASE_COMMIT_RE = re.compile(r"^Release \d+\.\d+\.\d+$")
-
-
 def get_previous_tag() -> str | None:
     result = subprocess.run(
         ["git", "describe", "--tags", "--abbrev=0", "--match", "v*", "HEAD^"],
@@ -37,26 +39,28 @@ def get_previous_tag() -> str | None:
     return tag if result.returncode == 0 and tag else None
 
 
-def get_commits_since(prev_tag: str | None) -> list[str]:
-    revision = f"{prev_tag}..HEAD" if prev_tag else "HEAD"
-    result = subprocess.run(
-        ["git", "log", revision, "--pretty=format:%s"],
-        capture_output=True,
-        text=True,
-        check=True,
-    )
-    lines = [line for line in result.stdout.splitlines() if line.strip()]
-    return [line for line in lines if not RELEASE_COMMIT_RE.match(line)]
+def changelog_section(text: str, version: str) -> str:
+    """Body of the ``## <version>`` section, up to the next ``## `` heading."""
+    heading = re.compile(rf"^## \[?{re.escape(version)}\]?(?:\s|$)", re.MULTILINE)
+    match = heading.search(text)
+    if match is None:
+        raise SystemExit(f"CHANGELOG.md has no section for {version}")
+    line_end = text.find("\n", match.start())
+    start = len(text) if line_end == -1 else line_end + 1
+    end = re.compile(r"^## ", re.MULTILINE).search(text, start)
+    body = text[start : end.start() if end else len(text)].strip()
+    if not body:
+        raise SystemExit(f"CHANGELOG.md section for {version} is empty")
+    return body
 
 
 def build_comment(
     tag_name: str,
-    commits: list[str],
+    notes: str,
     prev_tag: str | None,
     *,
     preview: bool = True,
 ) -> str:
-    items = "\n".join(f"- {c}" for c in commits) if commits else "- No changes."
     range_str = f"{prev_tag}...{tag_name}" if prev_tag else tag_name
     heading = "Changelog preview for" if preview else "Changelog for"
     blurb = (
@@ -67,22 +71,22 @@ def build_comment(
     return (
         f"## {heading} `{tag_name}`\n\n"
         f"{blurb}"
-        f"**What's Changed**\n\n{items}\n\n"
+        f"{notes}\n\n"
         f"**Full Changelog**: {range_str}\n"
     )
 
 
 def main() -> None:
     output = os.environ.get("OUTPUT", "/tmp/changelog-preview.md")
-    # release.yml sets TAG_NAME on tag builds; release-pr.yml never does, so its
+    # release.yml sets TAG_NAME on tag builds and release-pr.yml never does. Its
     # absence is what distinguishes a PR preview from the real release notes.
     tag_name = os.environ.get("TAG_NAME")
     preview = not tag_name
     tag_name = tag_name or f"v{get_version()}"
 
+    notes = changelog_section(CHANGELOG.read_text(), tag_name.removeprefix("v"))
     prev_tag = get_previous_tag()
-    commits = get_commits_since(prev_tag)
-    Path(output).write_text(build_comment(tag_name, commits, prev_tag, preview=preview))
+    Path(output).write_text(build_comment(tag_name, notes, prev_tag, preview=preview))
 
 
 if __name__ == "__main__":

tests/test_build_changelog_comment.py

@@ -3,9 +3,34 @@
 import importlib.util
 from pathlib import Path
 
+import pytest
+
 SCRIPT = Path(__file__).parent.parent / "scripts" / "build_changelog_comment.py"
 
-COMMITS = ["Add a thing (#1)", "Fix another thing (#2)"]
+CHANGELOG = """# Changelog
+
+Intro text.
+
+## Unreleased
+
+- Not released yet.
+
+## 1.2.3 - 2026-10-21
+
+### Added
+
+- A thing.
+
+### Fixed
+
+- Another thing.
+
+## 1.2.2 - 2026-09-30
+
+### Fixed
+
+- An older thing.
+"""
 
 
 def load_script():
@@ -15,58 +40,74 @@ def load_script():
     return module
 
 
+def test_notes_are_the_version_section_of_the_changelog():
+    module = load_script()
+    notes = module.changelog_section(CHANGELOG, "1.2.3")
+
+    assert notes == "### Added\n\n- A thing.\n\n### Fixed\n\n- Another thing."
+
+
+def test_last_section_runs_to_the_end_of_the_file():
+    module = load_script()
+    assert (
+        module.changelog_section(CHANGELOG, "1.2.2") == "### Fixed\n\n- An older thing."
+    )
+
+
+def test_bracketed_version_headings_are_found():
+    module = load_script()
+    text = "## [2.0.0] - 2026-11-01\n\n- New.\n"
+    assert module.changelog_section(text, "2.0.0") == "- New."
+
+
+def test_version_prefix_does_not_match_a_longer_version():
+    module = load_script()
+    with pytest.raises(SystemExit, match=r"no section for 1\.2$"):
+        module.changelog_section(CHANGELOG, "1.2")
+
+
+def test_missing_or_empty_section_fails_the_build():
+    module = load_script()
+    with pytest.raises(SystemExit, match=r"no section for 9\.9\.9"):
+        module.changelog_section(CHANGELOG, "9.9.9")
+    with pytest.raises(SystemExit, match="is empty"):
+        module.changelog_section("## 1.0.0\n\n## 0.9.0\n\n- Old.\n", "1.0.0")
+
+
 def test_preview_wording_on_pull_requests():
     module = load_script()
-    text = module.build_comment("v1.2.3", COMMITS, "v1.2.2", preview=True)
+    text = module.build_comment("v1.2.3", "- A thing.", "v1.2.2", preview=True)
 
     assert text.startswith("## Changelog preview for `v1.2.3`")
     assert "> Preview of the release notes" in text
+    assert "- A thing." in text
     assert "**Full Changelog**: v1.2.2...v1.2.3" in text
-    for commit in COMMITS:
-        assert f"- {commit}" in text
 
 
 def test_release_notes_drop_the_preview_wording():
     """Tag builds render the notes themselves, not a preview of them."""
     module = load_script()
-    text = module.build_comment("v1.2.3", COMMITS, "v1.2.2", preview=False)
+    text = module.build_comment("v1.2.3", "- A thing.", "v1.2.2", preview=False)
 
     assert text.startswith("## Changelog for `v1.2.3`")
     assert "preview" not in text.lower()
     assert "**Full Changelog**: v1.2.2...v1.2.3" in text
-    for commit in COMMITS:
-        assert f"- {commit}" in text
 
 
 def test_first_release_falls_back_to_the_tag_alone():
     module = load_script()
-    text = module.build_comment("v0.1.0", COMMITS, None, preview=False)
+    text = module.build_comment("v0.1.0", "- First.", None, preview=False)
 
     assert "**Full Changelog**: v0.1.0" in text
 
 
-def test_empty_commit_range_is_reported():
-    module = load_script()
-    text = module.build_comment("v1.2.3", [], "v1.2.2", preview=False)
-
-    assert "- No changes." in text
-
-
-def test_release_commits_are_filtered_from_the_log(monkeypatch):
+def test_get_version_reads_pyproject_without_tomllib():
     module = load_script()
-    log = "Release 1.2.3\nAdd a thing (#1)\n\nBump version to 1.2.3\n"
-    monkeypatch.setattr(
-        module.subprocess,
-        "run",
-        lambda *a, **k: type("R", (), {"stdout": log, "returncode": 0})(),
-    )
-
-    assert module.get_commits_since("v1.2.2") == [
-        "Add a thing (#1)",
-        "Bump version to 1.2.3",
-    ]
+    assert module.get_version().count(".") >= 1
 
 
-def test_get_version_reads_pyproject_without_tomllib():
+def test_current_version_has_a_changelog_section():
+    """Catches a release whose CHANGELOG.md entries were left under Unreleased."""
     module = load_script()
-    assert module.get_version().count(".") >= 1
+    changelog = (SCRIPT.parent.parent / "CHANGELOG.md").read_text()
+    assert module.changelog_section(changelog, module.get_version())

tests/test_event_serialization.py

@@ -1,5 +1,6 @@
 from __future__ import annotations
 
+from wildedge import constants
 from wildedge.events import common
 from wildedge.events.error import ErrorCode, ErrorEvent
 from wildedge.events.feedback import FeedbackEvent, FeedbackType
@@ -77,6 +78,33 @@ def test_feedback_event_enum_and_string_forms():
     assert string_event.to_dict()["feedback"]["feedback_type"] == "reject"
 
 
+def test_error_event_caps_message_and_provider_code():
+    event = ErrorEvent(
+        model_id="m1",
+        error_code=ErrorCode.UNKNOWN,
+        error_message="x" * 5000,
+        provider_error_code="c" * 500,
+    ).to_dict()["error"]
+    assert len(event["error_message"]) == constants.ERROR_MSG_MAX_LEN
+    assert len(event["provider_error_code"]) == constants.PROVIDER_ERROR_CODE_MAX_LEN
+
+
+def test_long_error_message_keeps_start_and_end():
+    message = "ffmpeg banner " * 100 + "Error opening input file audio.wav"
+    sent = ErrorEvent(
+        model_id="m1", error_code=ErrorCode.UNKNOWN, error_message=message
+    ).to_dict()["error"]["error_message"]
+    assert len(sent) == constants.ERROR_MSG_MAX_LEN
+    assert sent.startswith("ffmpeg banner")
+    assert sent.endswith("Error opening input file audio.wav")
+    assert " … " in sent
+
+
+def test_short_error_message_is_unchanged():
+    event = ErrorEvent(model_id="m1", error_code="UNKNOWN", error_message="boom")
+    assert event.to_dict()["error"]["error_message"] == "boom"
+
+
 def test_span_event_to_dict_includes_required_fields():
     event = SpanEvent(
         kind="tool",

wildedge/constants.py

@@ -53,7 +53,7 @@
 IDLE_POLL_INTERVAL = 5.0
 
 # Error message truncation
-ERROR_MSG_MAX_LEN = 200
+ERROR_MSG_MAX_LEN = 500
 PROVIDER_ERROR_CODE_MAX_LEN = 64
 
 # Device ID persistence

wildedge/decorators.py

@@ -4,7 +4,6 @@
 import time
 from typing import TYPE_CHECKING, Any
 
-from wildedge import constants
 from wildedge.timing import elapsed_ms
 
 if TYPE_CHECKING:
@@ -87,7 +86,7 @@ def wrapper(*args, **kwargs):
                 if self.capture_errors:
                     self.handle.track_error(
                         error_code="UNKNOWN",
-                        error_message=str(exc)[: constants.ERROR_MSG_MAX_LEN],
+                        error_message=str(exc),
                         **self._correlation,
                     )
                 raise
@@ -106,9 +105,7 @@ def __exit__(self, exc_type, exc_val, exc_tb):
             if self.capture_errors:
                 self.handle.track_error(
                     error_code="UNKNOWN",
-                    error_message=str(exc_val)[: constants.ERROR_MSG_MAX_LEN]
-                    if exc_val
-                    else None,
+                    error_message=str(exc_val) if exc_val else None,
                     **self._correlation,
                 )
         else:

wildedge/events/error.py

@@ -6,6 +6,7 @@
 from enum import Enum
 from typing import Any
 
+from wildedge import constants
 from wildedge.events.common import add_optional_fields
 
 
@@ -19,6 +20,29 @@ class ErrorCode(str, Enum):
     UNKNOWN = "UNKNOWN"
 
 
+ELLIPSIS = " … "
+
+
+def truncate(value: str | None, limit: int) -> str | None:
+    return None if value is None else str(value)[:limit]
+
+
+def truncate_message(value: str | None, limit: int) -> str | None:
+    """Keep the start and the end of a long message.
+
+    Errors put their detail at either end. mlx-whisper prints the ffmpeg banner
+    first and the failure last, and Hugging Face Hub starts with a request id.
+    """
+    if value is None:
+        return None
+    value = str(value)
+    if len(value) <= limit:
+        return value
+    keep = limit - len(ELLIPSIS)
+    head = keep - keep // 2
+    return value[:head] + ELLIPSIS + value[len(value) - keep // 2 :]
+
+
 @dataclass
 class ErrorEvent:
     model_id: str
@@ -49,11 +73,17 @@ def to_dict(self) -> dict:
         error_data = add_optional_fields(
             {"error_code": code},
             {
-                "error_message": self.error_message,
+                # Capped here, so every error event gets the same limit,
+                # from an integration or a manual track_error() call.
+                "error_message": truncate_message(
+                    self.error_message, constants.ERROR_MSG_MAX_LEN
+                ),
                 "stack_trace_hash": self.stack_trace_hash,
                 "related_event_id": self.related_event_id,
                 "http_status": self.http_status,
-                "provider_error_code": self.provider_error_code,
+                "provider_error_code": truncate(
+                    self.provider_error_code, constants.PROVIDER_ERROR_CODE_MAX_LEN
+                ),
                 "duration_ms": self.duration_ms,
             },
         )