Repository navigation
Conversation
|
Why exactly do you change it to a standalone executable ? |
2c36fff to
22d264b
Compare
|
Good question. The standalone executable was an intermediate attempt, not the final approach. The inline custom build step was broken in three ways: After fixing the inline step instead — |
The build-time header preprocessor (which generates c/loadable-ext-*.h for loadable extensions) was broken in three ways: 1. readOriginalData used readAlloc, which always failed against the 0.16/0.17 Io.Reader API. 2. processor.dump() was never called, so the output header was just a comment banner. 3. the consumer path was not wired into the build, so zig build test never exercised it. Fix the inline PreprocessStep to use readFileAlloc, call dump(), and depend on the generated headers. Bump SQLite to 3.53.4, regenerate the committed c/loadable-ext-*.h headers, and update the CI matrix.
aa12c3b to
b49eecd
Compare
The header preprocessor (which generates c/loadable-ext-*.h) was broken since the 0.16 Io.
Reader migration — readAlloc always failed, dump() was never called, and CI never exercised the consumer path.
Rewrite as a standalone host executable that takes paths via stdin.
Also bumps SQLite to 3.53.4, regenerates the committed headers, and fixes the CI matrix.
Depends on PR2 (feat/compat-0.16-0.17) for the cross-compile legs.