| | 57 | |
| | 58 | Update: I tried the following changes (patches attached): |
| | 59 | - Modify `dumper.py` so it sorts the library paths after uniquing them (1) and does not modify the global `options.library_paths` (2) |
| | 60 | - Modify `shlibs.py` so a path that does not start with `/` is considered a build path, and ignored (3) |
| | 61 | |
| | 62 | I then rebuilt from source all the packages, with the following results: |
| | 63 | - `g-ir-scanner` executes successfully on all packages |
| | 64 | - It generates the same files as before, except on `shared-library` attributes, as expected |
| | 65 | - All the incorrect relative paths `./directory/libsomething.dylib` are now `/opt/local/lib/libsomething.dylib` and the libraries actually exist at the specified paths. |
| | 66 | |
| | 67 | The only two remaining issues are: |
| | 68 | - The incorrect path `libavahi-core.so.7` in the `AvahiCore-0.6.gir` from package `avahi`. Looking at the package source, it seems that this `.gir` file is not generated by `g-ir-scanner` but is part of the sources, so this can only be fixed by modifying the `avahi` package. |
| | 69 | - The incorrect `.gir` files from the `webkit2-gtk` package come from the fact that the library path given to `g-ir-scanner` is |
| | 70 | {{{ |
| | 71 | /opt/local/var/macports/build/.../webkit2-gtk/work/build/lib |
| | 72 | }}} |
| | 73 | but the current working directory is |
| | 74 | {{{ |
| | 75 | /opt/local/var/macports/build/.../webkit2-gtk/work/build/Source/WebKit |
| | 76 | }}} |
| | 77 | Since it’s not a prefix of the library path, the library path is not detected as a build path and is used as the final library path in the `.gir` file. |
| | 78 | This could probably be fixed by changing the heuristic used by `g-ir-scanner` (use the last library path instead of the first? It seems to be `/opt/local/lib` more consistently) but could break other packages. |
