Code-health suggestion (proactive — staged for your review)
Category: hygiene — repository weight
Where: samples/v8/libv8_monolith.a (tracked at HEAD), plus samples/v8/include/libv8_monolith.a
(same archive at its pre-#112 path, still in history)
90% of this public repo is a prebuilt third-party static library, in two copies
Two blobs, both reachable from origin/main:
$ git ls-tree -r -l HEAD | awk '{s+=$4} END {printf "%d bytes / %d files\n", s, NR}'
66806257 bytes / 546 files
$ git ls-tree -r -l HEAD | sort -k4 -nr | head -3
100644 blob 6170aa8f… 62396638 samples/v8/libv8_monolith.a
100644 blob 87b36a29… 285431 clangwalk/krapped/clangwalk.cc
100644 blob 2d294ced… 145685 krapper/.../codegen/KotlinWriter.kt62,396,638 of 66,806,257 tracked bytes = 93.4% of the working tree, in one file. The
second-largest tracked file is 219× smaller.
Packed (what a clone actually downloads), over the blobs reachable from origin/main:
$ git rev-list origin/main --objects | awk '{print $1}' \
| git cat-file --batch-check='%(objectsize:disk) %(objecttype)' \
| awk '$2=="blob"{s+=$1;n++} END {printf "%d bytes across %d blobs\n", s, n}'
29052248 bytes across 2291 blobs
$ # the top two, by packed size, of those 2291:
14106228 6170aa8f… samples/v8/libv8_monolith.a (62396638 raw, at HEAD)
12082470 921760b3… samples/v8/include/libv8_monolith.a (55918284 raw, history only)
88988 2cb8ed1b… (the third-largest blob in the repo)26,188,698 of 29,052,248 packed blob bytes = 90.1%, and the gap to the third-largest blob is 147×.
gh repo view Monkopedia/kplusplus reports diskUsage 28,542 KB, which agrees.
The second copy is the same archive at the path it had before
61273a5 cleanup: unify samples under samples/ (#112) moved it. Git stores both; the move did not
free the old one. It is not LFS — grep -c filter=lfs samples/v8/.gitattributes → 0, and the only
rule in that file is the *.bat CRLF line.
What it is, and what depends on it
It is a build output of a third-party project — V8's libv8_monolith.a, consumed by
samples/v8/build.gradle.kts:139 (library("libv8_monolith.a")). Nothing in the repo produces it,
nothing pins the V8 revision it came from, and it carries no provenance or licence note of its own
beyond the V8 headers checked in beside it.
samples/v8 is a standalone build: not in settings.gradle.kts, in no workflow, last touched
2026-07-27 (b787196, the #184 tool merge — a mechanical rename, not use). So every clone of this
public repo pays ~26 MB for a sample that no gate here builds.
Proposed — this one is a decision, not a mechanical fix
Three options, in increasing cost and effect. (a) is the only one that shrinks an existing clone,
and it is a history rewrite, which is why this is filed for your call rather than dispatched.
- (a) Rewrite it out of history (
git filter-repo --path samples/v8/libv8_monolith.a --path samples/v8/include/libv8_monolith.a --invert-paths) and have samples/v8/README document how to
obtain or build the archive. Recovers ~26 MB of the 28 MB repo. Rewrites every commit SHA — every
open PR, fork and local clone has to re-base, and the issue tracker's commit links go stale. Needs
your explicit go-ahead.
- (b) Git LFS for
samples/v8/*.a. Keeps SHAs, but LFS objects still count against the account
and a plain clone no longer yields a working sample — and history keeps both existing copies
unless (a) is also done, so on its own it recovers nothing.
- (c) Untrack it going forward (
git rm --cached + .gitignore) and document the fetch step.
Costs nothing, breaks nobody, and recovers zero bytes from history — it only stops the number
growing when the archive is next refreshed. Honest floor, not a fix.
There is also a prior question that may make this moot: if samples/v8 is retired (it builds in
no workflow, and #231 already records that docs/ARCHITECTURE.md:64-72 and README.md:111-118 list
it while omitting samples/multiproject), then (a) becomes "delete the sample and rewrite the
archive out", which is one decision instead of two.
Why: this is the single largest thing in the repo by an order of magnitude, it is a third-party
binary with no provenance, and it is paid for by every clone of a public repository for a sample that
nothing here builds. It is also the kind of item that only gets cheaper to fix the earlier it is
decided — a rewrite gets more disruptive with every fork and open PR.
Effort: larger — (a) is irreversible and coordinated; (c) is mechanical but recovers nothing.
Refs #231, Refs #112.
Filed by the nightly code-health pass. To act on it: relabel agent-workable (it'll flow into triage→work). To decline: close it — it won't be suggested again.
Code-health suggestion (proactive — staged for your review)
Category: hygiene — repository weight
Where:
samples/v8/libv8_monolith.a(tracked at HEAD), plussamples/v8/include/libv8_monolith.a(same archive at its pre-#112 path, still in history)
90% of this public repo is a prebuilt third-party static library, in two copies
Two blobs, both reachable from
origin/main:62,396,638 of 66,806,257 tracked bytes = 93.4% of the working tree, in one file. The
second-largest tracked file is 219× smaller.
Packed (what a clone actually downloads), over the blobs reachable from
origin/main:26,188,698 of 29,052,248 packed blob bytes = 90.1%, and the gap to the third-largest blob is 147×.
gh repo view Monkopedia/kplusplusreportsdiskUsage28,542 KB, which agrees.The second copy is the same archive at the path it had before
61273a5 cleanup: unify samples under samples/ (#112)moved it. Git stores both; the move did notfree the old one. It is not LFS —
grep -c filter=lfs samples/v8/.gitattributes→0, and the onlyrule in that file is the
*.batCRLF line.What it is, and what depends on it
It is a build output of a third-party project — V8's
libv8_monolith.a, consumed bysamples/v8/build.gradle.kts:139(library("libv8_monolith.a")). Nothing in the repo produces it,nothing pins the V8 revision it came from, and it carries no provenance or licence note of its own
beyond the V8 headers checked in beside it.
samples/v8is a standalone build: not insettings.gradle.kts, in no workflow, last touched2026-07-27 (
b787196, the#184tool merge — a mechanical rename, not use). So every clone of thispublic repo pays ~26 MB for a sample that no gate here builds.
Proposed — this one is a decision, not a mechanical fix
Three options, in increasing cost and effect. (a) is the only one that shrinks an existing clone,
and it is a history rewrite, which is why this is filed for your call rather than dispatched.
git filter-repo --path samples/v8/libv8_monolith.a --path samples/v8/include/libv8_monolith.a --invert-paths) and havesamples/v8/READMEdocument how toobtain or build the archive. Recovers ~26 MB of the 28 MB repo. Rewrites every commit SHA — every
open PR, fork and local clone has to re-base, and the issue tracker's commit links go stale. Needs
your explicit go-ahead.
samples/v8/*.a. Keeps SHAs, but LFS objects still count against the accountand a plain
cloneno longer yields a working sample — and history keeps both existing copiesunless (a) is also done, so on its own it recovers nothing.
git rm --cached+.gitignore) and document the fetch step.Costs nothing, breaks nobody, and recovers zero bytes from history — it only stops the number
growing when the archive is next refreshed. Honest floor, not a fix.
There is also a prior question that may make this moot: if
samples/v8is retired (it builds inno workflow, and #231 already records that
docs/ARCHITECTURE.md:64-72andREADME.md:111-118listit while omitting
samples/multiproject), then (a) becomes "delete the sample and rewrite thearchive out", which is one decision instead of two.
Why: this is the single largest thing in the repo by an order of magnitude, it is a third-party
binary with no provenance, and it is paid for by every clone of a public repository for a sample that
nothing here builds. It is also the kind of item that only gets cheaper to fix the earlier it is
decided — a rewrite gets more disruptive with every fork and open PR.
Effort: larger — (a) is irreversible and coordinated; (c) is mechanical but recovers nothing.
Refs #231, Refs #112.
Filed by the nightly code-health pass. To act on it: relabel
agent-workable(it'll flow into triage→work). To decline: close it — it won't be suggested again.