build: fix typos in NAT-PMP detection block of configure.ac

[JSanchezFDZ]

Summary

Fix two typos in the NAT-PMP detection block of configure.ac that cause --disable-wallet builds (and likely --without-natpmp requests on systems with libnatpmp available) to either link-fail or silently misbehave.

The bug

The dnl Enable NAT-PMP support block (lines 1569-1595 on develop) is a copy of the equivalent UPnP block immediately above it, but with two introduced typos:

1. Line 1571 — if test "$have_natpmp" = "yes"; then inverts the disable/enable branches. The matching UPnP version (line 1543) reads if test "$have_miniupnpc" = "no"; then. The block decides what to do when the library was not found; with the typo it instead enters that branch when the library was found, so the auto-detect path runs the disable path on a successful detection and vice versa.

2. Line 1578 — if test "$use_natpmp" != "no"l; then. The trailing l is a stray character. The shell evaluates "$use_natpmp" != "no"l which is always true (no value of $use_natpmp ever equals the literal string "no"l). The matching UPnP version (line 1550) correctly reads if test "$use_upnp" != "no"; then.

The compounded effect of the two typos: USE_NATPMP ends up AC_DEFINE'd (so mapport.cpp compiles in the NAT-PMP code path, since it uses #ifdef USE_NATPMP which only checks for defined-ness, not value) while at the same time NATPMP_LIBS ends up empty (so -lnatpmp is not on the link line). The result on --disable-wallet builds of develop HEAD (989ec2e):

$ ./configure --disable-bench --without-gui --disable-wallet
...
checking for natpmp.h... yes
checking for initnatpmp in -lnatpmp... yes
...
$ make raptoreumd
...
/usr/bin/ld: libraptoreum_server.a(libraptoreum_server_a-mapport.o):
  in function `ProcessNatpmp()':
./src/mapport.cpp:95: undefined reference to `readnatpmpresponseorretry'
./src/mapport.cpp:137: undefined reference to `sendnewportmappingrequest'
./src/mapport.cpp:64:  undefined reference to `sendpublicaddressrequest'
collect2: error: ld returned 1 exit status

(The reason --enable-wallet builds happen to escape the failure on this host is irrelevant link-order luck — the underlying bug is the same in both configurations.)

Fix

Make the NAT-PMP block mirror the UPnP block character-for-character (modulo the rename miniupnpc → natpmp / upnp → natpmp):

• "yes" → "no" on the outer if test.
• Drop the stray l after "no" on the inner if test.
• (Cosmetic) determine → determines in the AC_DEFINE_UNQUOTED description string, matching the UPnP equivalent.

Test plan

Verified on develop HEAD (989ec2e) with build prerequisites: ./autogen.sh && CONFIG_SITE=depends/x86_64-pc-linux-gnu/share/config.site.

• Before patch, ./configure --disable-bench --without-gui --disable-wallet:

  • raptoreum-config.h contains #define USE_NATPMP 0 (a defined-but-zero value).
  • Makefile has NATPMP_LIBS = (empty).
  • make raptoreumd fails with the link errors quoted above.

• After patch, same configure:

  • raptoreum-config.h contains /* #undef USE_NATPMP */.
  • make raptoreumd builds successfully. Daemon starts, --help works.

• After patch, ./configure ... --enable-wallet --with-incompatible-bdb: still builds and runs cleanly. The --with-natpmp / --without-natpmp flags behave as documented.

Why this matters

• CI configurations and downstream packagers that build with --disable-wallet (headless nodes, masternode-only builds, future Guix-deterministic builds) hit the link failure today on develop.
• The typo means the auto-detect path silently does the wrong thing even when both flags are left at their defaults — depending on link order and which TU references which natpmp symbol first, the resulting binary may either fail to link, compile out NAT-PMP support when the user wanted it, or compile it in when they didn't.
• Pure-textual fix, no behaviour change to working configurations.

Related

• EVM integration — 32 commits across Phases 0–5 (draft for review) #443 (EVM integration) — surfaced this during full-suite build verification.
• test(amount): fix GetFeeTest int64 overflow causing 318-failure cascade #444 (amount_tests int64 overflow) and build: add ENABLE_WALLET guards to rpc/rpcassets.cpp #446 (ENABLE_WALLET guards in rpc/rpcassets.cpp) — two other regressions blocking the same --disable-wallet verification path.
• CFeeRate internal arithmetic overflows int64_t for fees near MAX_MONEY #445 — CFeeRate hardening tracking ticket.