Robert Sachunsky

> Ok, then (besides reformulation of the description) I highly recommend **renaming** that option, e.g. `apply_binarization`: after all, it's not the **input** that must/can be binary, but the internal **step**...

Seems to have been fixed by @tboenig with the newest releases. So was this a problem with the gt-repo-scripts alone (and we can close here), or is there still something...

The problem is back with all the released bags after and including https://github.com/OCR-D/gt_structure_text/releases/tag/v1.4.3. The problem is not with the bagger though: The culprit is [this change](https://github.com/OCR-D/gt_structure_text/commit/ba8e83216c7be9fc99c27472dfd1acc42f0198ff) in the PAGE files....

But even if the files are repaired in the GT data repo: IMO the bagger really needs to address this: https://github.com/OCR-D/core/blob/3a69e65cf93e2e1503b6af8ed898026889c46b15/src/ocrd/workspace_bagger.py#L104 (We do rely on derived images like binarization in...

> 1. What will be the effect of throwing exceptions for C applications and other applications which depend on the C API? Frankly, I don't know. I suppose we could...

BTW, doing `abort()` directly is also worse than going via exception handlers, even for the CLI, because it does not call destructors (and thus, does not shut down streams and...

> That's the normal nature of fatal errors. As their name says, they are not vital. That statement does not make any sense to me. > Even without the few...

> > @kba, @bertsky, please open issues for reproducable examples of failing assertions. > > Did I miss the examples? @bertsky, could you please send me a link? @kba promised...

We do have lots of GT for this kind of material, btw.

Ok, I think I found the problem: In https://github.com/qurator-spk/eynollah/blob/a2359ea4c45e6eb3eb704d79b1b011889c3ec8db/src/eynollah/eynollah.py#L1932 the textline mask gets **intersected** with the textregion masks. But the latter will be much narrower around the edges, hence the...