build: generate the coverage report without exuberant-ctags

[filmil]

Fixes #81

Root cause

make cov runs build/showprogress.sh, which extracted the ICU function lists with:

all=$(ctags -x --c-kinds=fp $header_fullname | ...)

--c-kinds is specific to exuberant-ctags. Other ctags implementations reject it, and where no ctags is installed the report cannot be generated at all.

While confirming this, I found the script fails even earlier on a current system. Line 31 called icu-config, which ICU removed in favor of pkg-config and which distributions no longer ship:

$ which ctags icu-config
(neither found)
$ bash build/showprogress.sh
build/showprogress.sh: line 31: icu-config: command not found

So the script aborted on its first real command, before reaching the ctags call.

The change

build/icu_c_functions.awk replaces the ctags call. awk is in POSIX, so nothing extra has to be installed.

It keys on how ICU declares its C API rather than on parsing C. Every public function is introduced by a line holding U_EXPORT2, with the name either starting the next line:

U_CAPI int32_t U_EXPORT2
umsg_format(const UMessageFormat *fmt, ...);

or trailing on the same one:

U_CAPI int32_t U_EXPORT2 uloc_countAvailable(void);

Keying on U_EXPORT2 rather than on U_CAPI also covers the older U_STABLE, U_DRAFT and U_DEPRECATED spellings, which take the same shape. Function pointer typedefs use U_CALLCONV instead and are skipped. ICU's handful of C++ convenience wrappers, declared inline <Type> with the name on the next line, are still reported, because ctags reported them.

The other changes:

• Fall back to pkg-config when icu-config is absent. icu-config is still tried first, since a hand-built ICU installs it and it reports that installation, which is what build/Dockerfile.maint sets up. Without this the script cannot run at all on a current system, so the ctags fix could not be verified.
• Drop exuberant-ctags from build/Dockerfile.buildenv. gawk is already installed there.
• Truncate the report detail file. It was opened with >>, so a run following one that died partway through appended to the leftovers. Easy to hit, since the script currently always dies partway through.
• Error out when a header yields no functions, instead of recording 0 / 0. The extraction keys on how ICU spells its declarations, so a future change to that should be loud rather than silent.
• Drop the grep -v U_DEFINE filter. It existed because ctags reported U_DEFINE_LOCAL_OPEN_POINTER macro invocations as functions. The awk extraction does not pick them up (checked on ICU 74 and ICU 77). Keeping it would have been actively harmful: under set -o pipefail, grep -v exits 1 on empty input, which killed the script silently and made the new "no functions found" check unreachable.

Verification

Ran on Ubuntu 24.04, ICU 74.2, with neither ctags nor icu-config installed.

The awk extraction agrees with ctags exactly. Running the old script and the new script against the same headers produces byte-identical output over all 40 files in coverage/:

$ diff -r old/coverage new/coverage && echo "IDENTICAL"
IDENTICAL

Per header, against both ctags implementations (exuberant-ctags 5.9, universal-ctags 5.9, which agree with each other on all 18):

18 headers, 471 symbols, mismatches: 0

Also checked against 10 ICU 77 headers, to confirm the extraction is not overfitted to the ICU version installed here. It matches ctags on 362 symbols, with three differences, all of which are ctags misparses that the awk extraction correctly skips:

Predicate   ucol.h:1547  explicit Predicate(const UCollator* ucol) : collator(ucol) {}
match       ucol.h:1594  bool match(std::u16string_view lhs, std::u16string_view rhs) const {
operator () ucol.h:1554  bool operator()(const T& lhs, const U& rhs) const {

These are C++ class members in a U_SHOW_CPLUSPLUS_API block, not ICU C API. This is why coverage/report.md changes in this PR: the checked-in report was generated against a newer ICU and lists match, operator and Predicate as unimplemented ICU functions. ucol.h goes from 8 / 54 to 8 / 51. That is the only content change to the checked-in report; no real function appears or disappears.

Checked with mawk 1.3.4, the default awk on Debian and Ubuntu. gawk and busybox awk were not installed here, so I did not run them; the script uses no extension beyond POSIX awk (match, substr, sub, RSTART, RLENGTH).

Error paths:

$ PKG_CONFIG_LIBDIR=/nonexistent PATH=/usr/bin:/bin bash build/showprogress.sh
error: cannot find the ICU headers: icu-config is not installed and pkg-config
does not know about icu-uc.  Install libicu-dev or the equivalent for your system.
exit=1

$ # include path pointing somewhere with no ICU headers
error: cannot read /nonexistent/unicode/ubrk.h

$ # header present but declaring nothing
error: no functions found in .../unicode/ubrk.h; build/icu_c_functions.awk needs updating

Workspace check:

$ cargo check --workspace
    Checking rust_icu v5.7.0 (/home/user/wt-81/rust_icu)
    Finished `dev` profile [unoptimized + debuginfo] target(s) in 30.08s

No crate sources changed, so there is no affected crate to run cargo test against. Nothing under .github/workflows/ is touched.

---

Generated by Claude Code

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.

[filmil]

The google-cla remarks is incorrect.