Roadmap

go-synctex/synctex is grown test-first, each capability covered before it ships. The parser — the source↔PDF query core an editor or viewer needs — is complete; the items below the line are honest "could come next", not promises.

Stage	What	Status
Stream parser	Line-oriented scan of the SyncTeX stream: the Input: file-tag map, the Content: marker, {P / } page blocks, Postamble:, and the [, h, v, k, g, x record lines with their N,L:X,Y prefix. Malformed lines skipped, not fatal.	Done
Gzip + plain input	Reads .synctex.gz (gzip) and plain .synctex, selected on the file suffix.	Done
Forward query	(source file, line) → nearest Record (page + x/y in sp), binary-searched over a per-source line index built at parse time.	Done
Backward query	(page, x, y in sp) → (source file, line) of the nearest record on the page (smallest squared distance).	Done
Source resolution	ResolveSource maps an editor-side project-relative path to the absolute path SyncTeX recorded (exact / suffix / base-name match).	Done
Project-root sanitisation	WithProjectRoot drops Input: paths outside the root and rewrites inside paths relative-to-root, resolving symlinks first — so host paths never leak to a client.	Done
Zip-bomb hardening	Decompressed body read through a LimitReader and rejected past a 32 MB cap before scanning.	Done
Tests & coverage	100% coverage including every error and malformed-record branch, gofmt + go vet clean, green across all six 64-bit Go arches (amd64, arm64, riscv64, loong64, ppc64le, s390x).	Done

Possible future work

These are candidates, not commitments — recorded so the surface's current boundaries are explicit:

• Typed sp / point helpers. A small ScaledPoint type (or ToPDFPoints helper) so callers don't hand-divide by 65536. Today the raw sp are exposed deliberately and the conversion is left to the caller; a typed convenience layer could sit on top without changing that contract.
• Box-dimension parsing (W,H,D). The parser reads the N,L:X,Y anchor and stops; the trailing width/height/depth are skipped because the current queries don't need them. Parsing them would enable box-extent-aware hit testing (e.g. "which box contains this click" rather than "which anchor is nearest").
• Streaming API. Parse reads the (capped) body into memory before scanning. A streaming variant that scans an io.Reader incrementally could lower the peak footprint for callers that already control the source.

Documented out-of-scope boundaries

• Coordinates are reported in raw scaled points. The parser does not apply Magnification: / Unit: / X Offset: / Y Offset: transforms; callers that need them apply them at the boundary. See The .synctex.gz format.
• The W,H,D box dimensions are not currently parsed — only the anchor X,Y. See the future-work note above.

See Usage & API for the surface and The .synctex.gz format for the stream these stages build on.