[Kartikey1306]

Stacked on #112. Review the last commit.

The default build does not compile

$ cmake -S . -B build && cmake --build build
net/src/net_posix.c:43:30: error: unknown type name 'eos_net_addr_t'
net/src/net_posix.c:73:1:  error: unknown type name 'eos_socket_t'
... 30 more

This is what the library job in CI — eos runs, and it is failing on master for this reason.

Why

eos/net.h puts every eNet declaration behind #if EOS_ENABLE_NET. CMakeLists.txt defines that flag only under:

if(EOS_BUILD_TESTS AND NOT EOS_PRODUCT)
    add_compile_definitions(... EOS_ENABLE_NET=1 ...)

but net/src/net_posix.c is in eos_net's source list unconditionally. Built without the feature, the header contributes no types and every definition in the file fails.

That is also why it stays invisible locally: -DEOS_BUILD_TESTS=ON is what anyone running the suite passes, and it happens to be the one configuration that defines the flag.

Fix

Guard the body the way the header is guarded, so the translation unit compiles away when the feature is off — what a disabled feature is supposed to do — rather than failing to compile.

Verified in both configurations

$ cmake -S . -B build && cmake --build build          # was: 32 errors
default build OK

$ cmake -S . -B build -DEOS_BUILD_TESTS=ON && ctest
100% tests passed, 34 of 34

$ nm libeos_net.a | grep -c eos_net_socket
1                                                      # still linked when the feature is on

[Kartikey1306]

Closing this in favour of #116, which covers the same two failures. Both PRs came from this account within a few hours — my fault for not checking first, and I'd rather leave one PR on the bug than two.

net_posix.c — functionally identical. Both add #if EOS_ENABLE_NET around the body; only the placement relative to the system includes and the comment wording differ. Nothing to choose between them.

test_net.c — this is why #116 rather than this one. This PR adds #include <assert.h>, which does fix the link error. #116 converts the ten call sites to the CHECK() macro the file already defines, and the file's own comment explains why that distinction matters:

Not CHECK(): CI builds Release, and NDEBUG deletes CHECK() together with the expression inside it. Half these checks wrap the call under test — CHECK(eos_net_connect(...) == 0) stopped connecting at all, the echo thread blocked in accept() forever, and pthread_join() hung the suite. A check macro must always evaluate its expression.

Adding the include makes those ten work only while add_compile_options(-UNDEBUG) stays in tests/CMakeLists.txt. If that line is ever moved, scoped differently, or a target is built outside it, assert() goes back to being deleted along with assert(eos_net_connect(...) == 0) — and the suite goes back to hanging in accept() rather than failing. That is the exact defect the CHECK() macro was added for, and four of the ten sites here are eos_net_connect / eos_net_send / eos_net_recv calls.

#116 also carries the verification I could not do from this branch: the default-configuration build reproduced against unpatched origin/master as a negative control, plus medical/aerospace profiles and ctest 34/34.

Nothing in this PR is lost.