The current digest hashes all of pyconfig.h, which means autoconf-cosmetic changes (_COSMO_SOURCE, future feature-test macros) invalidate the digest even though they don't reach stencil codegen.
Ideally, we could find a way to narrow the digest to inputs that actually affect codegen. Perhaps we could hash preprocessor output of stencil sources or similar?
Related to #148646 and #148647
The current digest hashes all of
pyconfig.h, which means autoconf-cosmetic changes (_COSMO_SOURCE, future feature-test macros) invalidate the digest even though they don't reach stencil codegen.Ideally, we could find a way to narrow the digest to inputs that actually affect codegen. Perhaps we could hash preprocessor output of stencil sources or similar?
Related to #148646 and #148647