perl: add support for loongarch64

[Cyanoxygen]

📦 Package Details

Maintainer: @jw2013

Description:
This PR allows Perl to be built for LoongArch64, as it lacks the configuration file. Both glibc and musl are tested using perl's t/base/*.t scripts.

---

🧪 Run Testing Details

• OpenWrt Version: master 7b7cf1454abe88463560f4c9463e578baab4276a
• OpenWrt Target/Subtarget: loongarch64/generic
• OpenWrt Device: QEMU, 3B6000

---

✅ Formalities

• I have reviewed the CONTRIBUTING.md file for detailed contributing guidelines.

[openwrt-ai]

Reviewed 1 new commit. The file name matches ARCH:=loongarch64 in target/linux/loongarch64/Makefile, so files/$(ARCH).config picks it up, and the PKG_RELEASE bump to 7 is right for a change with no PKG_VERSION update. The commit message matches what the diff does.

Four findings, all inline, all about symbols this config leaves unset compared with the eleven existing per-arch configs. The SIGCLD omission is the one I'd call a real defect — t/base/*.t would not catch it. The d_long_double_style_ieee / fpossize ones affect what lands in the generated config.sh; the ccsymbols one is a nit and I've phrased it as a question. None of them look like merge blockers to a maintainer's eye, but the signal list is worth fixing before merge.

---

Generated by Claude Code

[openwrt-ai]

SIGCLD is missing from the extra-signal list, so $SIG{CLD}, kill 'CLD', ... and POSIX's name↔number lookup won't know about it on loongarch64.

LoongArch uses the asm-generic signal numbering, where SIGCHLD is 17 and both glibc and musl alias SIGCLD to it. Every other arch config in this directory that uses asm-generic numbering lists it — e.g. riscv64.config:20-21 and aarch64.config:19-20 both use 'IOT CLD POLL UNUSED' / '6 17 29 31'. The two lists are consumed pairwise by signal.config, so both need the entry added at the same position.

Suggested change

	owrt:sig_name_extra='IOT POLL UNUSED'
	owrt:sig_num_extra='6 29 31'
	owrt:sig_name_extra='IOT CLD POLL UNUSED'
	owrt:sig_num_extra='6 17 29 31'

---

Generated by Claude Code

[openwrt-ai]

d_long_double_style_ieee='define' is set without any of the three concrete styles, which leaves the generated config.sh self-contradictory: base.config:308-311 defaults _doubledouble, _extended and _std all to undef, and this file overrides only the umbrella symbol.

Perl's Configure derives the umbrella from the specific one, and longdblkind='1' (IEEE 754 128-bit little endian, correct for LoongArch64's binary128 long double) maps to d_long_double_style_ieee_std. Compare x86_64.config:10-11, which sets d_long_double_style_ieee together with _extended for its longdblkind='3'.

Suggested change

	d_long_double_style_ieee='define'
	d_long_double_style_ieee='define'
	d_long_double_style_ieee_std='define'

---

Generated by Claude Code

[openwrt-ai]

fpossize is never defined for this target. It is set only in the per-arch config files — all eleven existing ones define it (fpossize='16' in riscv64.config:13, aarch64.config and x86_64.config; 20 or 24 elsewhere) — and it appears in neither base.config, architecture.config nor misc.config, so on loongarch64 it drops out of the generated config.sh and $Config{fpossize} ends up undefined. sizeof(fpos_t) is 16 on both glibc and musl for LP64.

Suggested change

	doublekind='3'
	doublekind='3'
	fpossize='16'

---

Generated by Claude Code

[openwrt-ai]

nit: ccsymbols, cppccsymbols and cppsymbols are omitted. Like fpossize, they are defined only in the per-arch files (see riscv64.config:5-7), and all eleven existing ones set at least the two empty ones — so on loongarch64 they will be absent from config.sh entirely rather than empty. Was this deliberate (the dumps are large and mostly informational), or just not captured from the reference build?

---

Generated by Claude Code