fix(json): only convert unambiguous date/datetime/UUID strings in additional data

[56badkarma]

Overview

JsonParseNode.try_get_anything runs every untyped string (everything landing in additional_data) through five lenient parsers in sequence: datetime_from_iso_format_compat, UUID, parse_timedelta_string, date.fromisoformat, time_from_iso_format_compat. On Python 3.11+ the lenient parsers accept many ordinary text values:

'19-2026'       -> datetime.time(19, 0, tzinfo=timezone(timedelta(days=-1, seconds=12840)))  # invoice number
'100-012863299' -> datetime.time(10, 0, ...)                                                 # invoice number
'PT'            -> datetime.timedelta(0)                                                     # country code
'10:30'         -> datetime.timedelta(seconds=37800)
'11.0'          -> datetime.time(11, 0)                                                      # version string
'd41d8cd98f00b204e9800998ecf8427e' -> UUID(...)                                              # 32-hex string

This is not just a display problem: in our production SharePoint AP-invoicing app, invoice numbers like 19-2026 were coerced to time objects, later stringified on update, and persisted back — permanently corrupting stored data (details with a captured wire payload in microsoftgraph/msgraph-sdk-python#1340).

This PR replaces the speculative parsing with two anchored patterns, mirroring the .NET JsonParseNode.TryGetAnything, which only attempts strict DateTime/DateTimeOffset/Guid parses and otherwise returns the string:

• full ISO date or datetime (^\d{4}-\d{2}-\d{2}([Tt ]\d{2}:\d{2}|$)) → datetime_from_iso_format_compat
• canonical hyphenated UUID → UUID

Behavior is unchanged for real payload values: full ISO datetimes still become datetime, date-only strings still become datetime (same as today, via the same compat parser), canonical UUIDs still become UUID. The timedelta/time guessing is removed entirely — Graph does not send bare durations or times as untyped strings, and their grammars overlap with too much real-world text. The __is_four_digit_number/isdigit() special cases (added for earlier symptom reports like microsoftgraph/msgraph-sdk-python#653) become unnecessary since all-digit strings cannot match either anchored pattern, and are removed.

Related Issue

Fixes #487
Addresses microsoftgraph/msgraph-sdk-python#1340 and microsoftgraph/msgraph-sdk-python#653

Notes

• Strings that previously coerced to timedelta/time/unhyphenated-UUID will now stay strings in additional_data. That is the intent of the fix, but callers who relied on those accidental conversions would see a change; typed model fields are unaffected (they go through declared field deserializers and never reach try_get_anything).
• Verified against a live Microsoft Graph SharePoint tenant: text columns round-trip unchanged, Created/Modified datetimes still parse.

Testing Instructions

• cd packages/serialization/json && pytest tests/
• New parametrized test test_get_anything_does_not_coerce_text_values covers each reported corruption pattern (19-2026, PT, 10:30, 23.085, 11.0, 32-hex, date fragments)
• New tests pin the preserved conversions: date-only → datetime, canonical UUID → UUID; existing tests cover full ISO datetimes

🤖 Generated with Claude Code

https://claude.ai/code/session_019Uui4g7E7UmzKRGHMW5zbH

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

[56badkarma]

@microsoft-github-policy-service agree