search

sirfz / tesserocr

A Python wrapper for the tesseract-ocr API
MIT License
2.02k stars 255 forks source link

add github-workflow building wheels #318

Closed betaboon closed 1 year ago

betaboon commented 1 year ago

resolves #123

Hello.

This PR adds a github-workflow to create wheels containing all the required libraries. builds include:

This now builds leptonica and tesseract from source.

~~some further information tho: since tesseract and leptonica are being installed via yum (for manylinux), apk (for musllinux) and brew (for macos) all of the built wheels contain a different version of tesseract. that might pose a problem to some. i think the only way to get around that would be to compile tesseract from source.~~

betaboon commented 1 year ago

@zdenop i think i addressed all your review comments now.

zdenop commented 1 year ago

I like your build scripts, but I am not sure if there should be part of tesserocr. Especially if not experienced user would use them. Here are some comments from perspective of tesserocr builder/user:

  1. On linux and Mac you should first check if tesseract (any probably leptonica) is not already installed. if yes then it should be uninstalled first before building tesseract/leptonica to avoid further problems.
  2. If your script should be used for creating tesserocr python package/wheel - you should stick to system provided leptonica/tesseract (API/ABI of different version are not the same and programs linked against shared libraries could be broken if different version is present). Other solution could be that you include your custom build shared libraries to tesseocr wheel as it is done for windows
  3. pango, cairo and icu4c (installed for MacOS) are needed only for training tools - tesserocr is not able to use them e.g. scripts (also for linux) should build tessaract without training tools if you build it only for tesserocr
  4. Personally I prefer to build leptonica with zlib and png support (AKA minimalistic build) - tesserocr is used for OCR, and python is able to open rest of image formats (with PIL or OpenCV). This decrease tesserocr dependency complexity...
zdenop commented 1 year ago

Regarding GitHub action part - this something I love to see in tesserocr (with windows wheels - but that is more difficult part).

betaboon commented 1 year ago

first off: thanks for taking the time to look at this.

I like your build scripts, but I am not sure if there should be part of tesserocr. Especially if not experienced user would use them.

as they are only intended to be used for the wheel builds in ci, we could move them to .github so that they are not considered for "public consumption".

1. On linux and Mac you should first check if tesseract (any probably leptonica) is not already installed. if yes then it should be uninstalled first before building tesseract/leptonica to avoid further problems.

2. If your script should be used for creating tesserocr python package/wheel - you should stick to system provided leptonica/tesseract (API/ABI of different version are not the same and programs linked against shared libraries could be broken if different version is present). Other solution could be that you include your custom build shared libraries to tesseocr wheel as [it is done for windows](https://github.com/simonflueckiger/tesserocr-windows_build/releases)

do i understand this correctly, that these points are only relevant when users would attempt to use the build-scripts locally?

3. `pango`,  `cairo` and  `icu4c` (installed for MacOS) are needed only for training tools - tesserocr is not able to use them e.g.  scripts (also for linux) should build tessaract without training tools if you build it only for tesserocr

do i get this right: you're suggesting to compile tesseract with -DBUILD_TRAINING_TOOLS=OFF and removing those dependencies?

4. Personally I prefer to build leptonica with zlib and png support (AKA minimalistic build) - tesserocr is used for OCR, and python is able to open rest of image formats (with PIL or OpenCV). This decrease tesserocr dependency complexity...

do i get this right: you're suggesting to to compile leptonic with -DENABLE_GIF=OFF, -DENABLE_JPEG=OFF, -DENABLE_TIFF=OFF, -DENABLE_WEBP=OFF and -DENABLE_OPENJPEG=OFF and removing those dependencies?

zdenop commented 1 year ago

For building release wheels use tesseract and leptonica provided by system.

If you plan to introduce testing CI (e.g. for commits, PR) then integrate those scripts to github actions as tesseract and leptonica do.

do i get this right: you're suggesting to compile tesseract with -DBUILD_TRAINING_TOOLS=OFF and removing those dependencies?

yes

do i get this right: you're suggesting to to compile leptonic with -DENABLE_GIF=OFF, -DENABLE_JPEG=OFF, -DENABLE_TIFF=OFF, -DENABLE_WEBP=OFF and -DENABLE_OPENJPEG=OFF and removing those dependencies?

yes.

betaboon commented 1 year ago

so i just pushed changes doing the following:

is there anything left for me to do here to get this merged?

betaboon commented 1 year ago

@sirfz thanks for merging.

quick question: when can we expect the wheels to be available on pypi?

sirfz commented 1 year ago

hi @betaboon, I'll include the wheels with the next release as I can't modify an existing release on pypi

betaboon commented 1 year ago

do you have any idea when that might be ?