codedb: byte-exact raw read (#632) + first-class `index` command (#633) by justrach · Pull Request #634 · justrach/codedb

justrach · 2026-06-21T09:44:58Z

Two codedb fixes surfaced while benchmarking codedb as an agent tool (claude -p / graff on SWE-bench), each filed with a failing zig test first (repo policy) then fixed in a separate commit. Full suite green; #632 also production-verified through the sandbox gateway on react (7,243 files).

#632 — `codedb_read` raw mode (byte-exact ranged read)

Ranged codedb_read emitted line-number-prefixed output (extractLines(line_numbers=true)) + a hash: header, so it was not a verbatim copy of the source — agents couldn't feed it to an exact-string editor and fell back to a native read (the codedb side of justrach/codegraff#66). Adds a raw arg: line_numbers=false, no hash header, no full-file hint. Default unchanged.

12c5555 failing test → 1009c64 fix. Closes codedb_read: ranged output is line-number-prefixed (not byte-exact) — add a raw read mode #632.

#633 — `index` is a first-class command

codedb <root> index triggered the scan/persist path (cold-load keys on cmd=="index") but wasn't in isCommand and had no dispatch branch, so it printed unknown command: index + exit 1 after doing the work; codedb index (no root) was a usage error. Registers index and adds a dispatch branch that reports index ready + exit 0.

2256739 failing test → 2c130a1 fix. Closes codedb index is a phantom command: scans + persists but exits 1 with 'unknown command: index' #633.

Base: release/0.2.5826.

…ber-prefixed, not byte-exact codedb_read with a line range emits extractLines(..., line_numbers=true) output plus a hash: header, so it is not a verbatim copy of the source. Agents that located code via codedb then fall back to a native read for the exact pre-edit span (see justrach/codegraff#66), so codedb never serves read+edit, only locate. The test asks for a raw/byte-exact ranged read (raw:true) and fails on main. Not fixing in this commit per repo policy. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Add a `raw` arg to codedb_read. In raw mode handleRead passes line_numbers=false to extractLines and suppresses the hash: header and the full-file read hint, so the body is a verbatim copy of the source — usable as the old_string for an exact-match edit. Default (prefixed, hash header) is unchanged. This unblocks routing in-repo reads through codedb (justrach/codegraff#66) so codedb can serve read+edit, not just locate. Closes #632. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

`codedb index` (no root) is a usage error and `codedb <root> index` falls through the dispatch to 'unknown command: index' (exit 1) even though the cold-load path already scanned + persisted. The test asserts parsePositional treats `index` as a first-class command; fails on main. No fix in this commit. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

`index` triggers the scan/persist path (the cold-load step keys on cmd==index) but was never registered in isCommand and had no dispatch branch, so it printed 'unknown command: index' + exit 1 after silently doing the work, and `codedb index` with no root was a usage error. Register it in isCommand and add an `index` dispatch branch that reports 'index ready' + exits 0. Closes #633. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

github-actions · 2026-06-21T09:47:56Z

Benchmark Regression Report

Thresholds: 10.00% and 50,000 ns absolute delta

NOISE means the percentage threshold was exceeded, but the absolute delta was too small to fail CI.

Tool	Base (ns)	Head (ns)	Delta	Abs Delta (ns)	Status
`codedb_bundle`	70710	73009	+3.25%	+2299	OK
`codedb_changes`	11914	10061	-15.55%	-1853	OK
`codedb_context`	754876	736967	-2.37%	-17909	OK
`codedb_deps`	326	320	-1.84%	-6	OK
`codedb_edit`	43465	40570	-6.66%	-2895	OK
`codedb_find`	3089	2893	-6.35%	-196	OK
`codedb_hot`	25728	23964	-6.86%	-1764	OK
`codedb_outline`	15473	15391	-0.53%	-82	OK
`codedb_read`	13169	12036	-8.60%	-1133	OK
`codedb_search`	61927	62278	+0.57%	+351	OK
`codedb_snapshot`	69698	72076	+3.41%	+2378	OK
`codedb_status`	12743	10682	-16.17%	-2061	OK
`codedb_symbol`	55711	57981	+4.07%	+2270	OK
`codedb_tree`	22743	22447	-1.30%	-296	OK
`codedb_word`	12005	11170	-6.96%	-835	OK

chatgpt-codex-connector

💡 Codex Review

codedb/src/mcp.zig

Lines 2988 to 2992 in 2c130a1

    
           if (explorer.renderCachedRead(path, alloc, out, .{ 
        
               .if_hash = if_hash, 
        
               .line_start = line_start_raw, 
        
               .line_end = line_end_raw, 
        
               .compact = compact,

Honor raw reads on cached content

When the file is still in Explorer.contents (the normal path for small MCP projects and freshly indexed files), this early renderCachedRead return runs before the new raw rendering below; ReadRenderOptions has no raw field, and renderReadBytes always emits the hash: header and numbered ranges. As a result, codedb_read with raw=true still returns non-raw output for cached files, so the new feature only works after contents have been released or on disk fallback.

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2026-06-21T09:48:01Z

        const end: u32 = if (line_end_raw) |n| @intCast(@min(@max(1, n), std.math.maxInt(u32))) else std.math.maxInt(u32);
        const lang = explore_mod.detectLanguage(path);
-        const extracted = explore_mod.extractLines(content, start, end, true, compact, lang, alloc) catch {
+        const extracted = explore_mod.extractLines(content, start, end, !raw, compact, lang, alloc) catch {


Preserve EOF without a final newline

For raw ranged reads that end at a file without a trailing newline, this still goes through extractLines, whose unnumbered path reprints every selected line with an added \n; for example, contents alpha\nbeta with line_start=1,line_end=2,raw=true returns alpha\nbeta\n. That extra byte makes the advertised byte-exact output unusable for exact-string edits at EOF, so raw mode should slice the original bytes for the requested range instead of formatting lines.

Useful? React with 👍 / 👎.

Add cases pinning the exact contract: raw full-file read is a byte-exact copy (no hash header, no line-number prefix, no full-file hint); the default ranged read still carries both the hash header and the 'N | ' prefix (regression guard); a raw ranged read drops both. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: d35804e6de

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2026-06-21T10:27:12Z

    const compact = getBool(args, "compact");
+    // #632: byte-exact ranged read — no line-number prefixes, no hash header — so
+    // the output can feed an exact-string editor instead of forcing a native read.
+    const raw = getBool(args, "raw");


Honor raw mode on cached reads

When the requested file is already in Explorer.contents (the normal post-scan MCP path, since indexed files are stored in the content cache), this new raw flag is ignored because renderCachedRead returns before the raw-aware rendering below and ReadRenderOptions has no raw field. In that scenario codedb_read with raw:true still emits the hash: header and N | line prefixes, so the byte-exact mode only works on the disk fallback path; the added tests miss this because they never index the file before reading it.

Useful? React with 👍 / 👎.

github-actions · 2026-06-21T10:28:00Z

Benchmark Regression Report

Thresholds: 10.00% and 50,000 ns absolute delta

NOISE means the percentage threshold was exceeded, but the absolute delta was too small to fail CI.

Tool	Base (ns)	Head (ns)	Delta	Abs Delta (ns)	Status
`codedb_bundle`	73765	76914	+4.27%	+3149	OK
`codedb_changes`	11144	11509	+3.28%	+365	OK
`codedb_context`	785754	760915	-3.16%	-24839	OK
`codedb_deps`	412	335	-18.69%	-77	OK
`codedb_edit`	41232	42616	+3.36%	+1384	OK
`codedb_find`	2905	2815	-3.10%	-90	OK
`codedb_hot`	25989	26122	+0.51%	+133	OK
`codedb_outline`	17671	16394	-7.23%	-1277	OK
`codedb_read`	14302	14204	-0.69%	-98	OK
`codedb_search`	67512	67348	-0.24%	-164	OK
`codedb_snapshot`	75107	74237	-1.16%	-870	OK
`codedb_status`	9632	10471	+8.71%	+839	OK
`codedb_symbol`	53730	55754	+3.77%	+2024	OK
`codedb_tree`	23225	25923	+11.62%	+2698	NOISE
`codedb_word`	12286	12126	-1.30%	-160	OK

justrach · 2026-06-22T12:30:20Z

Landed on release/0.2.5826 via fast-forward (rebased onto the post-split tip; full suite green). Commits 375989e..c72a129. GitHub won't badge this 'merged' since it didn't go through the merge button, but the patches are verifiably on the release tip.

justrach and others added 4 commits June 21, 2026 17:12

chatgpt-codex-connector Bot reviewed Jun 21, 2026

View reviewed changes

justrach closed this Jun 22, 2026

justrach deleted the fix/codedb-632-633 branch June 22, 2026 12:30

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

codedb: byte-exact raw read (#632) + first-class `index` command (#633)#634

codedb: byte-exact raw read (#632) + first-class `index` command (#633)#634
justrach wants to merge 5 commits into
release/0.2.5826from
fix/codedb-632-633

justrach commented Jun 21, 2026

Uh oh!

github-actions Bot commented Jun 21, 2026

Uh oh!

chatgpt-codex-connector Bot left a comment

Uh oh!

chatgpt-codex-connector Bot Jun 21, 2026

Uh oh!

chatgpt-codex-connector Bot left a comment

Uh oh!

chatgpt-codex-connector Bot Jun 21, 2026

Uh oh!

github-actions Bot commented Jun 21, 2026

Uh oh!

justrach commented Jun 22, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

	if (explorer.renderCachedRead(path, alloc, out, .{
	.if_hash = if_hash,
	.line_start = line_start_raw,
	.line_end = line_end_raw,
	.compact = compact,

Conversation

justrach commented Jun 21, 2026

#632 — codedb_read raw mode (byte-exact ranged read)

#633 — index is a first-class command

Uh oh!

github-actions Bot commented Jun 21, 2026

Benchmark Regression Report

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector Bot Jun 21, 2026

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector Bot Jun 21, 2026

Choose a reason for hiding this comment

Uh oh!

github-actions Bot commented Jun 21, 2026

Benchmark Regression Report

Uh oh!

justrach commented Jun 22, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

#632 — `codedb_read` raw mode (byte-exact ranged read)

#633 — `index` is a first-class command