You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
shell(cp): accept -f/--force, -r/--recursive, --verbose and --no-clobber - #47933
Each option spelling in that file is pinned here. The older tests have -R. The table has -r, -rf, -Rf, -fR, -rv, -vr, -Rv, -vR, -nrv, -frv, --recursive, --verbose and --force.
Commands that main accepts keep their result: -Rf, -vf, -Rv, -vR, -Rp, -RH, -RL, -RP, -Ri, -Rz, -Ra, -vH, -nz, -nf, -Rnf, -Rfn, -vnf, -n -Rf, -Rf -n. The tests pin them.
-n and -f
GNU cp 9.7, run here: -n -f, -f -n, -nf and -fn all keep the destination. BSD cp takes the last of the two. Bun's mv also takes the last one, as GNU mv and BSD mv do.
-n wins here. 1.4.4 already keeps the destination for cp -nf a b, because it ignores the f. The BSD rule would turn that command into an overwrite.
What -f does not do
POSIX and GNU -f remove a destination that cannot be opened and create it again. The builtin does not.
Linux, as a user that is not root: cp -f a b onto a mode 444 b prints cp: Permission denied: <path> and exits 1. cp a b and cp -Rf a b do the same on main. GNU cp 9.7 replaces b.
A copy keeps the mode of the source. So cp -f of a read-only source works one time and fails on the next run. GNU cp -f works each time.
macOS: a source above 128 KiB is cloned after the destination is unlinked, with or without -f. There the destination is replaced.
Windows: CopyFileW refuses a read-only destination with ACCESS_DENIED. That prints as cp: Operation not permitted.
The test -f copies onto a read-only file as plain cp does pins these results on each platform. uutils' test_cp_arg_force expects the replacement.
Letters that cp does not implement
Main ignores every letter after the first of a group. A reply on #46270 gives the reason: released versions exit 0 for cp -RL, -Rp, -Rf and -Ri, and an error would break scripts that work.
No released version accepts a group that starts with r or f. So the rule does not extend to them. cp -fi a b and cp -rT d e stay errors. So does a file name such as -fonts.css that cp *.css dist gets from the glob.
p gets no exception. cp -rp, cp -rfp and cp -fp stay errors, and they now name -p, as cp -pr and cp -r -p do. cp -Rp still works, because released versions accept it.
An earlier commit of this PR ignored a p after any letter, so that -rp worked as -Rp does. A review found that it exits 0 without the times that -p asks for. Such a choice is for a maintainer, so it is not in this PR.
For that choice: GitHub code search for "cp -rp" filename:package.json gives 258 files. 195 spell it cp -rp, 22 spell it cp -Rp, 24 are scp -rp. A test in that commit showed what a copy with an ignored p keeps: Windows keeps the time of a file (CopyFileW), Linux and macOS do not.
-a now prints unsupported option, as -p does. The same search gives 2,436 files for "cp -a" and 2,008 for "cp -f". -a is not made an alias of -R, because it also asks for -p.
Checked on every group of one to three letters from R r v n f p a H i L P T t u x z (4,368 groups, with a file pair and with a directory pair):
Main parses 819 groups, this PR 881. No group that main parses is rejected.
The 62 new groups start with r or f and hold only R r v n f.
For a group that main parses, the result with files never changes. The result with a directory changes for 60 groups. Each has an r after the first letter. It now copies the tree, as its -R twin does.
After a leading R, v or n, a letter is still ignored, as on 1.4.4. Two effects stay:
cp -vrT src dir now copies to dir/src. On main it exits 1, because the r was ignored too.
A file name that starts with -R, -v or -n is still read as options. With this PR, so is a file name that starts with -r or -f and holds only R r v n f. So is a file named --force, --recursive, --verbose or --no-clobber. Where the glob gives -f first, cp *f t with the files -f and kf copies only kf and exits 0. Main refuses that command, and GNU cp 9.7 does as this PR does. ./* prevents that. shell: accept -- end-of-options delimiter in builtins #33995 adds --.
Results that -r and -f inherit
-R and plain cp give these results on main. The new spellings reach them. Run on Linux against GNU cp 9.7. Each row has an open PR. All of these PRs conflict with main now.
Release builds (LTO) of main b9f2ec3 and of this PR, linux-x64. Tools: size, nm, gdb 16.3 (stepi, catch syscall). strace, perf, valgrind and bloaty are not installed here.
Size: .text 110,623,657 and .rodata 26,772,460 bytes on both builds. Cp::start goes from 2,164 to 1,813 bytes. parse_short is now a function of its own (259 bytes), because parse_long also calls it. No other symbol changes size.
Instructions of the main thread, from the entry of Cp::start until it returns. Second cp of one script, 3 runs each, spread 0:
Command
main
this PR
cp a b
855
860
cp -R d e
893
918
cp -n a b
892
918
cp -v a b
893
917
cp -Rf d e
911
941
cp -Rp d e
911
950
cp -Rvn d e
918
965
cp -f a b
916
cp -r d e
918
cp --force a b
925
cp --recursive d e
935
cp --verbose a b
930
cp --no-clobber a b
937
Allocator calls in that window: 7 in every row on both builds.
File and descriptor syscalls of all threads, from the second Cp::start to exit, 3 runs each, identical:
New spelling
Existing spelling
Syscalls
cp -r d e, cp --recursive d e
cp -R d e
23 each
cp --verbose a b
cp -v a b
19 each
cp --no-clobber a b, b exists
cp -n a b
9 each
cp -f a b, cp --force a b
cp a b
18 each, with and without b
The existing spellings give the same counts on main.
Debug and ASAN build of main b9f2ec3: 4 fail, 11 pass. With this PR: 15 pass.
Of the 71 table rows, 26 give the same result on main. They pin released behaviour. The other 45 exit 1 on main: 25 now exit 0, and 20 exit 1 with another message.
Mutations, each built and run:
the guard for every leader: 11 rows fail
no a arm: 3 rows fail
no v in the leaders: 1 row fails
f that resets -n: 6 rows and the read-only test fail
All 44 files of test/js/bun/shell and test/cli/run/run-shell.test.ts: 1,055 pass, 2 fail. The same 2 fail on main on this machine: shell load > immediate exit, and bun exec > it works2 when the whole directory runs in one process.
Windows and macOS do not run on my machine, so their expectations come from the code. The file passes on those lanes: 43 tests on Windows x64 and aarch64, 14 on macOS x64 and aarch64. Linux: 15.
The read-only test takes the unprivileged spawn options one time for its six commands. Two holders change the mode of the temporary root under each other.
On macOS the verbose lines of a directory copy depend on clonefile. The table accepts no line (the clone worked) or both lines (it did not).
no test proof · iteration 4 · platform-specific test(s) that do not run on this machine, deferring to CI, which covers all platforms: test/js/bun/shell/commands/cp.test.ts
The cp builtin rejected `-f` as the first flag of a group, had no arm for
`-r`, and took no long option.
`-f` is now accepted and sets nothing: the copy already overwrites unless
`-n` is given, and `-n` wins over `-f` in any order, as in GNU cp. `-r`
means `-R` in any position. `--recursive`, `--verbose`, `--no-clobber` and
`--force` run the arm of their letter.
A flag that cp does not implement is still ignored after a leading `R`,
`v` or `n`, as in the released versions. After `r` or `f` it is an error,
except `p`, which is ignored after any flag. `-p` and `-a` name themselves
in the unsupported-option error.
Fixes#14595
The diff is green. At 05fbdc8, test/js/bun/shell/commands/cp.test.ts passes on every lane of build 137686: Linux (15 tests), Windows x64 and aarch64 (43), macOS x64 and aarch64 (14). The build is red because of 27 files in test/aot/ and test/bundler/compile-aot-size.test.ts. They also fail on main, and this diff does not touch them.
…only test
Two concurrent tests each took and released `unprivilegedSpawnOptions`.
That helper changes the modes of the directories above the temporary
directory, so one test could remove the access that the child of the
other test needed. One test now runs the six commands under one call.
A second test pins the time of a file after `cp -fp`: the `p` is ignored,
and only the copy of Windows keeps the time.
We couldn't safely recover the incremental review. No full review was started, and the last reviewed checkpoint was preserved. Retry later, or explicitly request a full review by commenting @coderabbitai full review.
You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.
Use the checkbox below for a quick retry:
Trigger review
No actionable comments were generated in the recent review.
Walkthrough
The cp builtin now accepts additional short and long option forms, including lowercase -r. Tests cover option parsing, recursive copying, and copies to read-only destinations.
Changes
cp Option Parsing and Copy Behavior
Layer / File(s)
Summary
cp option parsing src/runtime/shell/builtin/cp.rs
The option documentation lists long forms. Parsing adds long-option mappings, accepts lowercase -r, treats -f as a no-op, and changes handling of grouped flags and unsupported -p.
Option parsing and recursive-copy tests test/js/bun/shell/commands/cp.test.ts
Child-process tests cover option forms and combinations, errors, output, copied files, and recursive copying from TypeScript and shell scripts.
Tests cover plain cp, -f, and -n -f on read-only destinations at two file sizes. The test helper forwards optional spawn options for unprivileged test cases.
Suggested reviewers: jarred-sumner
Severity of issue fixed: Medium
Merge Risk: Minimal · up to 05fbd
No confirmed merge-blocking risk remains in the reviewed changes.
Pre-merge checks: 4
Check name
Status
Explanation
Title check
The title names the main change: the Bun Shell cp builtin now accepts the -f/--force, -r/--recursive, --verbose and --no-clobber options. The raw summary and objectives confirm this is t…
Linked Issues check
Issue #14595 requires cp -r folder dest to work in Bun Shell on Windows, as it does in bash. At the reviewed head, Opts::parse_short in src/runtime/shell/builtin/cp.rs maps b'r' and b'R' to …
Out of Scope Changes check
The code change at the reviewed head is limited to the Opts option parser in src/runtime/shell/builtin/cp.rs. The PR description reports changes to cp.test.ts only. The parser changes go beyond …
Description check
The description explains what the PR changes, why the change is needed, known limitations, and how the code was verified. It does not use the exact template headings, but it provides the required info…
Full details: Title check
Explanation
The title names the main change: the Bun Shell cp builtin now accepts the -f/--force, -r/--recursive, --verbose and --no-clobber options. The raw summary and objectives confirm this is the primary change, and the title is concise and specific. It does not mention the -p error message change or the parsing leniency for later letters, but those are secondary details.
Full details: Linked Issues check
Explanation
Issue #14595 requires cp -r folder dest to work in Bun Shell on Windows, as it does in bash. At the reviewed head, Opts::parse_short in src/runtime/shell/builtin/cp.rs maps b'r' and b'R' to the same self.recursive = true state. The diff from the merge base confirms this change. The PR description states that the two #14595 repros are added as tests in test/js/bun/shell/commands/cp.test.ts. This assessment did not read the test file. The coding requirement is met in the parser. The test coverage claim is reported by the PR, not confirmed by a separate read. Windows behavior is not run here. The PR description says CI passed the test file on Windows x64 and aarch64.
Full details: Out of Scope Changes check
Explanation
The code change at the reviewed head is limited to the Opts option parser in src/runtime/shell/builtin/cp.rs. The PR description reports changes to cp.test.ts only. The parser changes go beyond the literal request in #14595, which covers -r only. They add -f as a no-op, the long forms --recursive, --verbose, --no-clobber and --force, -n precedence over -f, and new unsupported option results for -a and -p. These changes stay within the cp option parsing that the PR objectives describe. They do not touch other subsystems. The test helper change that forwards uid and gid supports the read-only destination tests. No unrelated change is shown.
Full details: Description check
Explanation
The description explains what the PR changes, why the change is needed, known limitations, and how the code was verified. It does not use the exact template headings, but it provides the required information in equivalent sections.
Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts
Reviewing files that changed from the base of the PR and between 7f66e25 and f3f8d01.
Files selected for processing: 2
src/runtime/shell/builtin/cp.rs
test/js/bun/shell/commands/cp.test.ts
Included review availability: This review used your included allowance. Your plan provides up to 10 included reviews per hour; 0 remain after this review.
The previous commit ignored a `p` that followed another flag, so
`cp -rp` and `cp -fp` exited 0 without the times that `-p` asks for.
They now fail with `unsupported option -- -p`, as `cp -pr` and
`cp -r -p` do. `cp -Rp` keeps working, because released versions
accept it.
The options test accepts both outputs of a verbose directory copy on
macOS: no line when clonefile copies the directory, and two lines when
that call fails.
The reason will be displayed to describe this comment to others. Learn more.
I re-reviewed after f3f8d01 and found no bugs; that push addresses both points from my earlier pass (-p after -r/-f now fails with unsupported option -- -p, and the macOS verbose directory row accepts either clonefile outcome). Because -f is accepted as a no-op rather than the POSIX unlink-and-recreate, and the "ignore later letters only after R/v/n" rule is a compatibility policy, a maintainer's call on those semantics is still worthwhile.
What was reviewed:
parse_long delegating to parse_short(short, flag, 0): the four long names can only hit the R/v/n/f arms, so the i > 0 and IllegalOption(&smallflags[i..]) arms are unreachable from that path.
The narrowed _ if i > 0 && matches!(smallflags[0], b'R' | b'v' | b'n') rule: every cluster the base accepted still parses; r/f-led clusters admit only R r v n f and otherwise fail loudly, which the table pins.
New tests: child pipes drained with Promise.all, temp dir and process released via using/await using, read-only destinations restored in finally, and unprivilegedSpawnOptions exists in test/harness.ts:829 with the Disposable shape the test uses.
Extended reasoning...
The diff touches only the FlagParser impl for Opts in src/runtime/shell/builtin/cp.rs (about 20 lines: long-name mapping, -r as -R, -f as a no-op, -a/-p error arms, and a narrowed ignore-after-first-letter rule) plus about 215 lines of new tests in test/js/bun/shell/commands/cp.test.ts. It touches no security-sensitive surface: no path handling, privilege, or network code changes, only option-letter dispatch. The latest commit resolves both optional findings from my previous review, no CODEOWNER covers these paths, and no other reviewer has requested changes. I deferred rather than approved because the change fixes user-facing shell compatibility policy (-f accepted without the POSIX force semantics, -Rp still silently ignoring p) that the PR itself says is a maintainer's choice.
`-Rv`, `-vR`, `-nrv` and `-frv` are the group orders in the tests of
#35616. `-RL` and `-RP` complete the groups that stay accepted after a
leading `R`.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
![claude[bot]](https://avatars.githubusercontent.com/in/1236702?s=60&v=4){kind=small}
Fixes #14595
Problem
cp -f a bprintscp: unsupported option, please open a GitHub issue -- -f.cp -r d eprintscp: illegal option -- r.impl FlagParser for Opts(src/runtime/shell/builtin/cp.rs:872) rejects a leadingfand has nor.Fix
-fis accepted and sets nothing.-rmeans-Rin any position. The title's long names map to their letters.-nis given, and-nwins over-f, as in GNU cp.test/js/bun/shell/commands/cp.test.ts(4 new tests fail on main). Alsotest/js/bun/shell.-rp(see Notes).Background
cpis a Bun Shell builtin, the onlycpon Windows. POSIX needsBUN_ENABLE_EXPERIMENTAL_SHELL_BUILTINS=1.R,vorn, because released versions accept-Rp. Afterrorfsuch a letter is an error.-fthat removes a destination it cannot open. It belongs in the copyfs.cpshares.-r/--recursive/--verboseand fix clustered short flags #35616. Merge before or with shell: keep the last exit status on ShellExecEnv and return it from a bareexit#47887.Downsides
cp -f a bonto abthat cannot be opened still fails, where GNU cp replacesb.cp -rpstays an error, because-pis not implemented.-rand-finherit every result of-Rand plaincp.cp -r dist/. out/fillsout/dist/(shell(cp): copy the contents of a dir/. or dir/.. source into the target #38067).cp -vrT src dircopies todir/src.cpwith an option, 5 without. No new allocation, syscall or binary byte.Notes
Where this comes from
cp -r folder deston Windows. Its two repros are tests here.exit#47887 makes a bareexitreturn the last status. Its review found package.json scripts of the shapecp -f a b || exit && next. On Windows they fail atcptoday and continue without the file. With shell: keep the last exit status on ShellExecEnv and return it from a bareexit#47887 alone they stop with status 1. With this PR the copy works.-r/--recursive/--verboseand fix clustered short flags #35616 had the same goal. It is written against the parser that Hardening follow-up: fix regressions from #46021 and more crashes #46270 replaced: its arms reject-Rpand drop the-nthat is now real. This PR replaces it.-r/--recursive/--verboseand fix clustered short flags #35616 runs unchanged on a build of this PR. Linux: 13 pass, 58 skip. With its Windows-only block forced on: 71 pass. On main, 9 of the 13 fail.-R. The table has-r,-rf,-Rf,-fR,-rv,-vr,-Rv,-vR,-nrv,-frv,--recursive,--verboseand--force.-r/--recursive/--verboseand fix clustered short flags #35616 makes-Rp,-RH,-RL,-RPand-Rierrors. Here they exit 0, as on main and on 1.4.3.What changes
Linux x64, builtin forced on, debug builds of main b9f2ec3 and of this PR. Every row exits 1 on main.
cp -f a b,-ff,-fv,-v -funsupported option ... -- -fcp -fR d e,cp -R -f d e,cp -frv d ecp -fn a b,cp -n -f a b,cp -f -n a bb, exit 0cp -r d e,-rR,-rf,-rvillegal option -- r(and the rest of the group)cp -vr d e,cp -nr d e,cp -nrv d ed is a directory (not copied)cp --force a b,--verbose a b,--recursive d eillegal option -- -force(and so on)-f,-v,-Rcp --no-clobber a billegal option -- -no-clobberb, exit 0cp -p a b,-pR,-R -punsupported option ... -- -P-- -pcp -rp d e,cp -rfp d e,cp -fp a billegal option -- rp,-- rfp,unsupported option ... -- -funsupported option ... -- -pcp -a d e,-ar,-raillegal option -- aunsupported option ... -- -acp -fi a b,cp -rL d e-- -f,illegal option -- rLunsupported option ... -- -i,-- -Lcp -rT d e,cp -ru a billegal option -- rT,-- ruillegal option -- T,-- ucp -f,cp -f aunsupported option ... -- -fCommands that main accepts keep their result:
-Rf,-vf,-Rv,-vR,-Rp,-RH,-RL,-RP,-Ri,-Rz,-Ra,-vH,-nz,-nf,-Rnf,-Rfn,-vnf,-n -Rf,-Rf -n. The tests pin them.-nand-f-n -f,-f -n,-nfand-fnall keep the destination. BSD cp takes the last of the two. Bun'smvalso takes the last one, as GNU mv and BSD mv do.-nwins here. 1.4.4 already keeps the destination forcp -nf a b, because it ignores thef. The BSD rule would turn that command into an overwrite.What
-fdoes not do-fremove a destination that cannot be opened and create it again. The builtin does not.cp -f a bonto a mode 444bprintscp: Permission denied: <path>and exits 1.cp a bandcp -Rf a bdo the same on main. GNU cp 9.7 replacesb.cp -fof a read-only source works one time and fails on the next run. GNUcp -fworks each time.-f. There the destination is replaced.CopyFileWrefuses a read-only destination withACCESS_DENIED. That prints ascp: Operation not permitted.cp_open_dest_with_mkdirand in theCopyFileWarm ofcopy_single_file_sync(src/runtime/node/node_fs.rs).fs.cpshares both. No open PR has it: draft node:fs: replace an existing destination inside the native single-file cp #38226 givesfs.cpa replace policy and keeps the shell in place.-f copies onto a read-only file as plain cp doespins these results on each platform. uutils'test_cp_arg_forceexpects the replacement.Letters that
cpdoes not implementcp -RL,-Rp,-Rfand-Ri, and an error would break scripts that work.rorf. So the rule does not extend to them.cp -fi a bandcp -rT d estay errors. So does a file name such as-fonts.cssthatcp *.css distgets from the glob.pgets no exception.cp -rp,cp -rfpandcp -fpstay errors, and they now name-p, ascp -prandcp -r -pdo.cp -Rpstill works, because released versions accept it.pafter any letter, so that-rpworked as-Rpdoes. A review found that it exits 0 without the times that-pasks for. Such a choice is for a maintainer, so it is not in this PR."cp -rp" filename:package.jsongives 258 files. 195 spell itcp -rp, 22 spell itcp -Rp, 24 arescp -rp. A test in that commit showed what a copy with an ignoredpkeeps: Windows keeps the time of a file (CopyFileW), Linux and macOS do not.b'p' if i > 0 => {}, would accept-rp. After a release, the rule of Hardening follow-up: fix regressions from #46021 and more crashes #46270 keeps it at exit 0.-anow printsunsupported option, as-pdoes. The same search gives 2,436 files for"cp -a"and 2,008 for"cp -f".-ais not made an alias of-R, because it also asks for-p.R r v n f p a H i L P T t u x z(4,368 groups, with a file pair and with a directory pair):rorfand hold onlyR r v n f.rafter the first letter. It now copies the tree, as its-Rtwin does.R,vorn, a letter is still ignored, as on 1.4.4. Two effects stay:cp -vrT src dirnow copies todir/src. On main it exits 1, because therwas ignored too.-R,-vor-nis still read as options. With this PR, so is a file name that starts with-ror-fand holds onlyR r v n f. So is a file named--force,--recursive,--verboseor--no-clobber. Where the glob gives-ffirst,cp *f twith the files-fandkfcopies onlykfand exits 0. Main refuses that command, and GNU cp 9.7 does as this PR does../*prevents that. shell: accept--end-of-options delimiter in builtins #33995 adds--.Results that
-rand-finherit-Rand plaincpgive these results on main. The new spellings reach them. Run on Linux against GNU cp 9.7. Each row has an open PR. All of these PRs conflict with main now.cp -r dist/. out/,outexistsout/dist/index.htmlout/index.htmlcp -r dist/. ../index.htmlcp -r d o2,d/lflinks tofo2/lflinks to the absolute path ofd/fo2/lflinks tofcp -f link outoutis a link to the absolute targetoutis a regular filecp -f link out,outexistsoutkeeps its content, exit 0outis replacedcp -r d missing/parent/e,cp -f a missing/parent/bdir/.change of shell(cp): copy the contents of a dir/. or dir/.. source into the target #38067 or the exit status change of shell(cp): exit 1 when a held-back Windows error turns out to be real #37943. Each changes the result of a command that exits 0 on a release. shell(cp): copy the contents of a dir/. or dir/.. source into the target #38067 needs a re-cut for the currentrun_from_thread_pool_impl.Not in this PR
-a,-p,-H,-L,-Pand-istay unimplemented.c,l,s,Xandxfrom the usage line stayillegal option.--rec, and options after the operands. The builtin does not.rm --force(shell(rm): accept --force as the long form of -f #39224),mkdir --verbose(shell(mkdir): accept --verbose instead of the misspelling --vebose #39221),mv --force,ls --all.--(shell: accept--end-of-options delimiter in builtins #33995), a lone-(shell: treat a lone-as an operand in the builtin option parsers #39214),--name=value(shell: report --name=value spellings of unsupported builtin options as unsupported #39223), the text ofillegal optionerrors (shell: name the rejected flag in the builtins' illegal option errors #39230)..shfile in cp -r illegal option in bun shell on Windows #14595 has CRLF line ends. With LF the script works now. With CRLF the target becomesdest\r(Bun Shell on Windows fails to handle CRLF line endings in .sh files, causing command resolution failure #29669, PR Shell: handle CRLF, tab, comment, and backslash-newline line endings in the lexer #29670).Measurements
Release builds (LTO) of main b9f2ec3 and of this PR, linux-x64. Tools:
size,nm, gdb 16.3 (stepi,catch syscall). strace, perf, valgrind and bloaty are not installed here..text110,623,657 and.rodata26,772,460 bytes on both builds.Cp::startgoes from 2,164 to 1,813 bytes.parse_shortis now a function of its own (259 bytes), becauseparse_longalso calls it. No other symbol changes size.Cp::startuntil it returns. Secondcpof one script, 3 runs each, spread 0:cp a bcp -R d ecp -n a bcp -v a bcp -Rf d ecp -Rp d ecp -Rvn d ecp -f a bcp -r d ecp --force a bcp --recursive d ecp --verbose a bcp --no-clobber a bCp::startto exit, 3 runs each, identical:cp -r d e,cp --recursive d ecp -R d ecp --verbose a bcp -v a bcp --no-clobber a b,bexistscp -n a bcp -f a b,cp --force a bcp a bbTests
aarm: 3 rows failvin the leaders: 1 row failsfthat resets-n: 6 rows and the read-only test failtest/js/bun/shellandtest/cli/run/run-shell.test.ts: 1,055 pass, 2 fail. The same 2 fail on main on this machine:shell load > immediate exit, andbun exec > it works2when the whole directory runs in one process.clonefile. The table accepts no line (the clone worked) or both lines (it did not).no test proof · iteration 4 · platform-specific test(s) that do not run on this machine, deferring to CI, which covers all platforms: test/js/bun/shell/commands/cp.test.ts