Source: rust-leptonica-sys Followup-For: Bug #1088237 X-Debbugs-Cc: noisyc...@tutanota.com Control: forwarded -1 https://github.com/ccouzens/leptonica-sys/issues/29
Having spent some more time debugging this issue, I now see I looked at the wrong autopkgtest run for leptonica-sys. The one I linked to in the initial report was for v0.4.8-3, which depends on the updated bindgen, so obviously layout tests were not run as regular tests but as compile-time tests (thus reducing the number of executed tests from 73 to 3) as expected. On the other hand, v0.4.8-2 autopkgtests *do* execute bindgen layout tests and *do* fail on i386, see e.g. [1]. The reason why this was not caught earlier, I think, is that the bug was introduced by an update in leptonlib: the first autopkgtest failure happened in early october [2], after v0.4.8-2 had already migrated to testing, on the first test run that used leptonlib 1.84.1-4. Indeed, it looks like upstream doesn't even support 1.84 yet, which may explain why even they didn't catch it (asides from, well, this bug being i686-specific). So it looks like all the bindgen update did was make the bug leak to other packages by turning autopkgtests to compile-time tests. I submitted an isse upstream hoping that this will be more helpful than mere investigative work. One temporary workaround while this gets resolved could be disabling layout tests on i386 [3] (if you are interested in this approach I could write a patch). Best. [1]: https://ci.debian.net/packages/r/rust-leptonica-sys/testing/i386/54616656/ [2]: https://ci.debian.net/packages/r/rust-leptonica-sys/testing/i386/52766436/ [3]: https://docs.rs/bindgen/0.70.1/bindgen/struct.Builder.html#method.layout_tests
