[fcostaoliveira]

What

Two opt-in options for script/amalgamate.py, plus a check script, so that the projects that vendor a flattened copy of the library can switch to the generated files:

• --with-log also emits the log codec: hdr_histogram_log.c, hdr_encoding.c, hdr_time.c and their public headers hdr_histogram_log.h, hdr_time.h. Each .c has its private headers inlined.
• --include-prefix hdr/ makes the generated .c include "hdr/hdr_histogram.h", for trees that keep the public headers in a hdr/ directory.
• script/check-amalgamation.sh verifies both, and is run by the amalgamation CI job and by the release job before anything is attached.

The release now also attaches hdr_histogram-amalgamation-with-log-<version>.zip. docs/AMALGAMATION.md gains a section on migrating an existing vendored copy.

No library source changes. Without the new flags the generator's output is byte-identical to what is on main (checked for the default and for --malloc-include).

Why

I tried replacing the vendored copies of Redis, Valkey, memtier_benchmark and Node.js with the generated files, now that #162, #164, #165, #166 and #167 are in. Two of the four did not work as is:

• memtier_benchmark also uses the log codec (hdr_log_*) and the time helpers. The log files include private headers (hdr_tests.h, hdr_histogram_internal.h, hdr_encoding.h, hdr_endian.h) that the release assets do not ship, plus <hdr/...> includes, so compiling them next to the core amalgamation failed on all three files.
• Node.js keeps src/ plus include/hdr/. The generated hdr_histogram.c includes "hdr_histogram.h", which does not resolve there: fatal error: 'hdr_histogram.h' file not found.

Redis and Valkey work with the existing options (Valkey's CMakeLists.txt also needs one line removed, since it lists hdr_atomic.h; that is in the new doc section).

Validation

• check-amalgamation.sh passes with gcc and clang under dash: each of the four amalgamated files preprocesses to the same translation unit as its source and builds with -Wall -Wextra -Wmissing-prototypes -Werror; a log encode/decode round trip works using only the generated files; --with-log --malloc-include routes the core and the codec through the caller's allocator; and with --include-prefix hdr/ the header is byte-identical to include/hdr/hdr_histogram.h and the file compiles with -Isrc -Iinclude. It fails (exit 1) when the generator is deliberately broken.
• Every run: step of both workflows was executed locally, including the new two-zip packaging and checksums.
• Redis (checkout @ 9af6e958d), replaced with no Makefile change: builds with 0 compiler warnings; the latency-monitor, info, redis-cli and redis-benchmark suites pass (122 tests). redis-benchmark histogram output steps by 0.1 ms up to 2.1 ms and by 1.0 ms after, as the iterator setter should.
• Valkey (an export of its tree): builds with 0 warnings; latency-monitor, info, valkey-benchmark and valkey-cli suites pass (142 tests).
• memtier_benchmark (@ 4ebfa71): replaced, builds from a fresh autoreconf/configure/make. The same workload (2 threads x 5 clients x 20,000 requests with --hdr-file-prefix and --json-out-file) through the old and the new build records 200,000 requests in both, and the compressed histograms written by hdr_log_encode decode correctly with the independent Python hdrh library. memtier's own integration tests were not run.
• Node.js (@ 463711aa): the generated file, in Node's layout and with histogram.gyp unchanged, compiles inside Node's own build with Node's flags, and Node's src/histogram.cc compiles against the replaced header. Node's code never stores a negative count (subtract refuses or clamps to 0, the CBOR reader accepts unsigned only), so it fits the non-negative counts contract. The full Node link and its test suite had not finished when this was opened; I will comment with the result.

Not tested

• The real gh release upload and the release: published trigger need an actual release. A manual "Run workflow" exercises everything except the upload.
• Only x86-64 Linux with gcc 13 and clang 18. The check script is not run on macOS or Windows.

🤖 Generated with Claude Code

[github-actions]

🤖 Automated first-pass review — a human maintainer's review is still required before merge.

The generator changes look fine, and the default/--malloc-include output stays the same. One thing to fix: check-amalgamation.sh builds with -D_GNU_SOURCE (CMake adds it on UNIX too, CMakeLists.txt:54). Without it, the amalgamated hdr_time.c won't compile under -std=c99 on glibc, because clock_gettime/CLOCK_MONOTONIC aren't declared. Either note that in the --with-log section of docs/AMALGAMATION.md, or drop the define from the check so it tests what an embedder actually gets. Smaller points: --with-log --include-prefix hdr/ together isn't exercised by the check, and the per-project migration recipes in the doc will go stale as Redis, Valkey, memtier and Node move on. I'd also wait for the Node link/test result you said you'd post.

[fcostaoliveira]

Node.js result, as promised in the description: a full build of nodejs/node @ 463711aa with deps/histogram replaced finished (./configure && make, gcc 13, x86-64 Linux) and the resulting node v27.0.0-pre runs.

What changed in deps/histogram: src/hdr_histogram.c and include/hdr/hdr_histogram.h replaced by the output of script/amalgamate.py --include-prefix hdr/, and src/hdr_atomic.h and src/hdr_tests.h deleted. histogram.gyp, BUILD.gn and unofficial.gni are untouched.

The histogram-related tests all pass: 21 run, 21 ok, 0 failed, 0 skipped (parallel/test-perf-hooks-histogram* including the diff, import, snapshot, stats, qrde, qrde-oracle, qrde-worker and record-zero cases, the sliding-window, timerify and monitor-event-loop-delay tests, and sequential/test-performance-eventloopdelay and test-perf-hooks-histogram-heapdump). The binary contains symbols that exist only in the new code (hdr_iter_linear_set_value_units_per_bucket, hdr_record_value_capped, hdr_total_count, the AVX2 scan), so the replacement is what was linked.

I ran only the histogram-related tests, not Node's whole suite.