fix(dbapi): avoid IndexError in get_operation_name for comment/whites…

[Atishyy27]

Description

get_operation_name in the DB-API instrumentation strips a leading SQL comment and
takes the first token as the operation name. If a statement is truthy but has no tokens
left after stripping (a comment-only or whitespace-only statement), .split()[0] raises
IndexError, which propagates out of the instrumented execute() call and breaks it.

This guards the result the same way the adjacent _Template branch already does
(tokens[0] if tokens else "") and returns an empty operation name instead. It extends the
empty-string handling added in #2643 to the comment-only / whitespace-only case.

No open issue for this one — found while reading the operation-name parsing. Related: #2643.

Type of change

• Bug fix (non-breaking change which fixes an issue)

How Has This Been Tested?

Added test_operation_name_no_tokens_after_comment_strip in test_dbapi_integration.py,
which executes a comment-only (/* comment only */) and a whitespace-only (" ")
statement and asserts instrumentation does not raise. Verified locally that it passes and
that the existing test_span_succeeded still passes.

Does This PR Require a Core Repo Change?

• No.

Checklist:

• Followed the style guidelines of this project
• Changelogs have been updated
• Unit tests have been added
• Documentation has been updated

[henry3260]

Thanks for the fix! Just some nits.
I think we should add a changelog fragment under .changelog/ instead of editing CHANGELOG.md directly.

[Atishyy27]

also moved the changelog to a .changelog/4934.fixed fragment per your note.

[chuenchen309]

Ran both this and #4936 against main at 038eb07ba. The failure reproduces, and the repo has already decided this case once, in a way that argues for both patches.

Reproduced. Calling get_operation_name directly, with a control group so the harness is honest:

dbapi.CursorTracer.get_operation_name (str branch)
  '/* comment only */'  -> IndexError: list index out of range
  '   '                 -> IndexError: list index out of range
  ''                    -> ''            (the #2643 guard)
  'SELECT 1'            -> 'SELECT'
  '/* c */ SELECT 1'    -> 'SELECT'

psycopg.CursorTracer.get_operation_name (Composable branch, #4936)
  same six inputs, same six results

get_operation_name is called unconditionally from _traced_execution, so this comes out of the instrumented execute() rather than being contained.

There is already a precedent in-tree, and it is the same fix. asyncpg hits the identical expression in three places (_do_execute, _do_cursor_execute, _do_prepared_execute) and each one is wrapped:

try:
    name = self._leading_comment_remover.sub("", name).split()[0]
except IndexError:
    name = ""

So "no tokens left after stripping ⇒ empty operation name" is already this repo's answer to this exact input; dbapi and psycopg are the packages that never got it.

Two sites are still uncovered after both PRs. Same probe, same clean checkout:

site	on main
psycopg2/__init__.py:335 — get_operation_name, str branch	raises on both inputs
sqlalchemy/engine.py:319 — _operation_name, parts.append(...)	raises on both inputs
asyncpg ×3	guarded already

psycopg2 is the same override shape #4936 fixes, one package over. Not asking you to widen these PRs — you already asked for one logical change each — but if you are doing them one at a time, that is the rest of the list.

One thing neither PR changes, mentioned so it is not mistaken for a gap. The regex is ^/\*.*?\*/: anchored, no re.DOTALL, and it does not know about -- comments. So '-- just a note\n' does not raise — it returns '--', which becomes the span name. Degenerate output rather than a crash, so it is a different question from this PR, but "empty after stripping" is not the only way that parse ends badly.

AI-assisted; every result above is output from running the code on 038eb07ba, not read off the source.

[opentelemetry-pr-dashboard]

Pull request dashboard status

Waiting on maintainers · refreshed 2026-08-21 18:39 UTC

Merge when ready.

Status above doesn't look right?

• Just replied or pushed? Anything around or after the refresh time above may not be picked up yet — give it a few minutes.
• Anything look wrong? Report it with what you expected; it helps us improve the dashboard.

[Atishyy27]

Addressed the review feedback and added the related fixes for psycopg, psycopg2, and SQLAlchemy. The PR is ready for maintainer review/merge.