Closed stefanCCS closed 1 year ago
We should probably add a clean
target to ocrd_anybaseocr and delegate to it on make clean
in ocrd_all, like we do for deps-ubuntu
. What do you think @stweil @bertsky
This is one of several nagging issues with the current build. Ideally running make
should not change any of the submodules after they were checked out or at least restore them after building. Currently some files are modified, others are added:
modified: cor-asv-ann (untracked content)
modified: dinglehopper (untracked content)
modified: ocrd_anybaseocr (modified content, untracked content)
modified: ocrd_olahd_client (untracked content)
modified: ocrd_tesserocr (modified content, untracked content)
modified: sbb_textline_detector (untracked content)
@stefanCCS, these commands should help:
git -C ocrd_anybaseocr/ocrd_anybaseocr/pix2pixhd reset --hard
git -C ocrd_anybaseocr/ocrd_anybaseocr/pix2pixhd clean -fxd
touch ocrd_anybaseocr
They restore changed files, remove added files and force a new build for ocrd_anybaseocr
.
@kba, this issue is one of several others which I also have observed where incremental builds fail to work.
In this particular case, I think we can keep ocrd_all completely out of the picture: It was the non-existing packaging of pix2pixHD and the horrible integration of the pix2pixHD subcomponent in ocrd_anybaseocr which led us to this solution. But now that we need a proper branch for that anyway, why not apply our packaging patches in another branch, then merge both in an OCR-D fork's master, and refer to that?
(@kba if we can agree on this plan then I recommend transferring the issue to ocrd_anybaseocr, which I have no permissions for.)
It was the non-existing packaging of pix2pixHD and the horrible integration of the pix2pixHD subcomponent in ocrd_anybaseocr which led us to this solution. But now that we need a proper branch for that anyway, why not apply our packaging patches in another branch, then merge both in an OCR-D fork's master, and refer to that?
In that vein, I have created https://github.com/NVIDIA/pix2pixHD/pull/297, so we can rebase ocrd_anybaseocr on that.
Alas, due to pypa/pip#6658, we cannot just make setup contain a pix2pixhd dependency as a relative path. So I think I'll just upload a pkg for it myself.
Done:
So I think I'll just upload a pkg for it myself.
(@kba if we can agree on this plan then I recommend transferring the issue to ocrd_anybaseocr, which I have no permissions for.
I could but you solved everything AFAICS. Is there still a need for it?
Currently some files are modified, others are added
pip install .
creates a build
directory in each submodule which uses that command. Some submodules already had an entry for that build artifact, but in several others that entry is missing. I therefore created several pull requests to fix that:
https://github.com/ASVLeipzig/cor-asv-ann/pull/10 https://github.com/OCR-D/ocrd_olahd_client/pull/4 https://github.com/qurator-spk/dinglehopper/pull/66 https://github.com/qurator-spk/sbb_textline_detection/pull/59
As soon as those pull requests were merged and ocrd_all
was updated to use the new code we should have a cleaner git status
after make all
.
I could but you solved everything AFAICS. Is there still a need for it?
No, not anymore.
@kba : Feel free to close this issue, as "make all" has worked already later in 2022.
Hi, I have tried to update the current (23.02.2022) ocrd_all
using git pull
andmake clean
andmake all
. Unfortunately, I get this error: