json-schema-to-grammar : report incomplete-conversion warnings only once

[RIVOIRA]

In llama-server the tool-call grammar is (re)built on every request (common/chat.cpp build_chat_peg_parser -> build_grammar), so a tool schema containing a regex pattern with no grammar equivalent (e.g. a lookahead, as in some agentic clients' tool definitions) logs a "WARNING: JSON schema conversion was incomplete" line on every request.

Report each distinct warning only once per process; repeats are suppressed. Behavior is unchanged otherwise - the affected fields still fall back to "any string".

Overview

Problem

llama-server logs one

WARNING: JSON schema conversion was incomplete: pattern … is not supported (unsupported group syntax), accepting any string

line per request whenever a tool schema contains a regex pattern with no
grammar equivalent (lookaheads (?!…) / (?=…) / (?<!…), named groups, …).
Such patterns appear in real agentic clients' tool definitions (path-segment
validation regexes), so a server serving one such client spams the log once
per turn — in a short session: 74 warning lines for 78 tasks.

Why it repeats

The tool-call grammar is (re)built on every request
(common/chat.cpp → build_chat_peg_parser → build_grammar()), and each
build_grammar() call creates a fresh common_schema_converter.
check_errors() therefore re-emits the same warning on every request.

Fix

Report each distinct warning once per process: check_errors() now keeps a
process-wide set of already-emitted warning strings (guarded by a mutex).
First occurrence still prints the usual WARNING line; repeats are suppressed.

No behavior change: the affected fields still fall back to "any string", and
the full diagnostic (with the offending pattern) is in the log from the first
occurrence.

A follow-up could cache the compiled grammar keyed on the tool schema to
avoid the per-request rebuild altogether; that is a much larger change and is
out of scope here.

Requirements

• I have read and agree with the contributing guidelines
• AI usage disclosure: YES — Claude Code was used to investigate the warning and draft the patch. I reviewed the change and verified it locally before submitting (unit test + functional before/after comparison).

[ggml-gh-bot]

Hi @RIVOIRA, thanks for your contribution!

Per our contribution guidelines, the automated PR checker found the following issue(s) that need your attention:

• AI-generated content: While code is allowed to be generated by AI, please write the PR description and commit messages on your own without the help of AI.

---

Please note that maintainers reserve the right to make final decisions on PRs. If you believe there is a mistake, please comment below.