[Stuartwastaken]

strip.py kept both sides of #ifndef __UNIT_TEST__ ... #else ... #endif: the #ifndef starts in the keep state and the #else set keep again instead of switching to strip. So the stripped homa_tx_pool.c defines homa_tx_pool_mem_nid twice (the real one and the unit-test stub) and doesn't compile.

The #else now flips the state. Only homa_tx_pool.c's stripped output changes (the stub is gone); every other file that make net-next copies strips the same as before. python3 util/strip.py homa_tx_pool.c | grep -c '^int homa_tx_pool_mem_nid' gives 2 before and 1 after.

[randomizedcoder]

We hit the same bug independently while building the make net-next files in a 7.3 tree (redefinition of 'homa_tx_pool_mem_nid') and came up with the identical fix, so we've dropped ours in favour of this one. The fix looks correct to us: the stripped output of every homa*.[ch] is unchanged except for the stub in homa_tx_pool.c. Together with an sk_memcg fix for kernels without CONFIG_MEMCG, the stripped files build cleanly in-tree (defconfig, W=1, gcc + sparse, clang, and i386). Thanks!