fix(openai): keep a cleared provider timeout from disabling every request

[Lesereingrape]

ProviderOpenAIOfficial hands the provider timeout field straight to the OpenAI SDK:

self.timeout = provider_config.get("timeout", 120)
self.custom_headers = self.request_headers
if isinstance(self.timeout, str):
    self.timeout = int(self.timeout)

timeout is a dashboard numeric field declared as a plain "type": "int" with no slider and no
minimum (astrbot/core/config/default.py:2822-2826), and clearing such a field writes 0 —
toNumber() maps parseFloat('') to 0
(dashboard/src/components/shared/ConfigItemRenderer.vue:361-364). httpx reads 0 as "expire
immediately", so every request of that provider dies before it leaves. A non-numeric value
("" from a cleared timeout: in a hand-edited data/config.json, or a quoted string) raises
ValueError inside __init__, which ProviderManager.initialize logs as a bare
Failed to load provider adapter ... Unknown cause. (astrbot/core/provider/manager.py:688)
before skipping the provider, so the model silently disappears from the selection.

Measured against a local OpenAI-compatible server that answers after 3 s, on master @ ac4df996:

provider_config["timeout"]	before this change	after this change
absent	client.timeout=120 → replies in 3.16 s	unchanged
120	replies in 3.02 s	unchanged
0	APITimeoutError: Request timed out. after 0.00 s	replies in 3.02 s
-5	APITimeoutError after 0.00 s	clamped to the shared 1 s floor, warns
""	ValueError: invalid literal for int() with base 10: '' → provider dropped	replies in 3.02 s
"abc"	same ValueError	replies in 3.02 s

The same __init__ is shared by 10 OpenAI-compatible adapters (groq, longcat, mirarouter,
aihubmix, openai_responses, openrouter, ssycloud, xai, xiaomi, zhipu), so one cleared box breaks
all of them.

Modifications / 改动点

• Route the field through coerce_int_config (astrbot/core/utils/config_number.py), the helper
  the repo already uses for exactly this class of field — including its own guard for the same
  timeout key in deerflow_agent_runner.py:197 — instead of adding a bespoke normalizer, so the
  behaviour matches the DeerFlow runner: a cleared, None or non-numeric value behaves like an
  absent one (documented default 120), a usable value is kept, and anything below the shared
  min_value=1 floor is clamped with a warning that names the field and the provider config.
• Add tests to the existing tests/test_openai_source.py covering the default, a tuned value, the
  numeric string, 0/negative/""/"abc"/None, and one OpenAI-compatible subclass inheriting
  the fallback.

No dependencies added, no signature or config schema change, no behaviour change for a valid
timeout.

Screenshots or Test Results / 运行截图或测试结果

Before the change (RED, the seven timeout cases against master @ ac4df996):

6 failed, 4 passed

After the change (GREEN):

10 passed
76 passed in 11.74s        # whole tests/test_openai_source.py

The gate CI runs for these paths, and the full suite:

ruff format --check astrbot/core/provider/sources/openai_source.py tests/test_openai_source.py   # 2 files already formatted
ruff check astrbot/core/provider/sources/openai_source.py tests/test_openai_source.py            # All checks passed!

3605 passed, 82 skipped in 453.99s        # full tests/ on master @ ac4df996 + this change

Two notes for the reviewer:

• .github/copilot-instructions.md:27 says "Do not generate test files for now." I did not add a
  file — the cases went into the existing tests/test_openai_source.py, which already tests this
  provider, and they are the measurement behind the table above. Say the word and I will drop them
  and leave the four-line change.
• fix: strip x-stainless-* telemetry headers from provider requests #9401 edits this file just above the timeout block (pristine 344-357 vs. 362-365 here), so the
  two changes touch disjoint lines; if fix: strip x-stainless-* telemetry headers from provider requests #9401 lands first this hunk just shifts.

检查清单未勾选，是因为这些第一条人称声明需由账号本人确认；本 PR 由自动化 agent 会话代为准备、测试并提交，实测证据已附于上文。

---

Disclosure: this PR was prepared, tested and submitted by an AI agent working on behalf of the account owner.

Summary by Sourcery

Ensure OpenAI-compatible providers resolve malformed or cleared timeout settings to a usable request timeout.

Bug Fixes:

• Prevent cleared, invalid, zero, and negative OpenAI provider timeout values from disabling requests or causing the provider to be dropped during initialization.

Enhancements:

• Apply shared integer configuration coercion with a default timeout and minimum supported value across OpenAI-compatible providers.

Tests:

• Add coverage for default, valid, string, invalid, empty, null, zero, negative, and inherited OpenAI-compatible provider timeout values.

[sourcery-ai]

Hey - I've reviewed your changes and they look great!

Sourcery assessment

Approved.

---

Sourcery is free for open source - if you like our reviews please consider sharing them ✨

• X
• Mastodon
• LinkedIn
• Facebook