"Version-Specific SQL Files" → "Don't `.gitignore` a Skipped Version — `rm` It Once" covers one case well: a version that's normally tracked but occasionally skipped, where the leftover generated file is transient (present for one build cycle) and the right cleanup is a one-time `rm`.
At least two consumer repos (cat_tools, extension_tools) use a different pattern this section doesn't address: a pseudo-version (e.g. `stable`) that a repo's `default_version` sits at permanently between releases, not a version that's skipped once. Unlike the skipped-version case, this file would be regenerated and re-diffed on every single source edit if tracked — it's always the current build target, never stable historical content — for zero test-coverage value (a fresh install from source is already tested on every push regardless). The correct handling is the opposite of the skipped-version guidance: gitignore it, not `rm` it once.
Worth a short subsection distinguishing this from the skipped-version case, so consumer repos don't have to independently arrive at (and re-justify) the same reasoning.
"Version-Specific SQL Files" → "Don't `.gitignore` a Skipped Version — `rm` It Once" covers one case well: a version that's normally tracked but occasionally skipped, where the leftover generated file is transient (present for one build cycle) and the right cleanup is a one-time `rm`.
At least two consumer repos (cat_tools, extension_tools) use a different pattern this section doesn't address: a pseudo-version (e.g. `stable`) that a repo's `default_version` sits at permanently between releases, not a version that's skipped once. Unlike the skipped-version case, this file would be regenerated and re-diffed on every single source edit if tracked — it's always the current build target, never stable historical content — for zero test-coverage value (a fresh install from source is already tested on every push regardless). The correct handling is the opposite of the skipped-version guidance: gitignore it, not `rm` it once.
Worth a short subsection distinguishing this from the skipped-version case, so consumer repos don't have to independently arrive at (and re-justify) the same reasoning.