Modernize codebase, add scaling, and fix memory leaks

[csingleplus]

Hello! I remembered this tool from quite some time ago, and a few months ago I noticed it needed adjustments to build correctly. This PR brings the original public tree closer to the state already maintained in my fork, with the addition of scaling.

I’m opening this both in case it is useful upstream and so users still landing on the original repository have a visible path to a maintained working version.

My fork currently maintains:

• a legacy branch for older systems
• a working master branch for current toolchains and the latest updates
• a provided .deb for users who want the easiest install path

This PR is based on the current development side of that work and includes:

• cleanup of dead disabled legacy code
• migration away from older Tesseract usage
• a fix for a small (67 byte) memory leak
• subtitle scaling/rendering improvements. more on the way!
• testing against multiple real-world DVD subtitle samples

Note: Most of these fixes are only implemented on the modern master branch. The legacy branch remains compatibility-oriented and does not include the newer code. Further work will include preprocessing of text, as sometimes, depending on the font in the subtitle, letters don't get recognized properly. While untested on them, the prebuilt modern .deb may work with systems from before ~2020.

Finally, I'd like to thank Ruediger Sonderfeld for creating this tool. It's been quite a good learning experience delving into old multimedia code like this, working through compiler errors and tracing Valgrind reports.

[milahu]

https://github.com/csingleplus/VobSub2SRT/blob/44b55903b19d72770566d375e7ee6048916410b6/CMakeLists.txt#L67

find_package(Tesseract REQUIRED)

should require tesseract >=4

with tesseract 3, it fails to compile with

/build/source/src/vobsub2srt.c++: In function 'int main(int, char**)':
/build/source/src/vobsub2srt.c++:280:53: error: 'OEM_LSTM_ONLY' is not a member of 'tesseract'; did you mean 'OEM_CUBE_ONLY'?
  280 |   if(tess_base_api.Init(NULL, tess_lang, tesseract::OEM_LSTM_ONLY) == -1) {
      |                                                     ^~~~~~~~~~~~~
      |                                                     OEM_CUBE_ONLY

---

with libpng 1.6.56, it fails to compile with

-- Found PNG: /nix/store/gsn3vddway3289p6mzy5shd1paly8dp4-libpng-apng-1.6.56/lib/libpng.so (found version "1.6.56")

/build/source/src/vobsub2srt.c++:30:10: fatal error: libpng/png.h: No such file or directory
   30 | #include <libpng/png.h>
      |          ^~~~~~~~~~~~~~

fixed by

-#include <libpng/png.h>
+#include <png.h>

---

without libtiff, cmake shows the cryptic error

CMake Error at /nix/store/v2i1hgv567g3v91im5x4g5bff52143i0-cmake-4.1.2/share/cmake-4.1/Modules/FindPackageHandleStandardArgs.cmake:227 (message):
  Could NOT find Tesseract (missing: Tesseract_LIBRARIES)
Call Stack (most recent call first):
  /nix/store/v2i1hgv567g3v91im5x4g5bff52143i0-cmake-4.1.2/share/cmake-4.1/Modules/FindPackageHandleStandardArgs.cmake:591 (_FPHSA_FAILURE_MESSAGE)
  CMakeModules/FindTesseract.cmake:57 (find_package_handle_standard_args)
  CMakeLists.txt:67 (find_package)

so maybe add

+find_package(TIFF REQUIRED) # for Tesseract
 find_package(Tesseract REQUIRED)

---

the language CLI parameter is not working
as soon as i set for example --lang deu i get garbage output in the SRT file
https://github.com/milahu/vobsub2srt-test-mad-max-1979-german

[csingleplus]

Seeing that you're on NixOS, I will install a VM with it and have a look at these issues straight away. Thanks for bringing them up!

Edit: I have tried your Mad Max .idx/.sub on my system, and it does recognize near perfect. So yes, I will have a look at this apparent NixOS-specific issue. Seems to do with a specific filesystem store Tesseract data may not be recognized normally.

[milahu]

I will install a VM with it

actually, you can install Nix on any Linux distro, and then run

git clone https://github.com/milahu/vobsub2srt-test-mad-max-1979-german
cd vobsub2srt-test-mad-max-1979-german
nix-shell
command -v vobsub2srt

(but of course a VM is better for sandboxing)

[csingleplus]

I've uploaded a complete SRT (some mis-identified letters, but mostly accurate) here:
https://github.com/csingleplus/VobSub2SRT/tree/master/test4milahu. Will start work on the fix very soon, and thank you for the info on Nix itself. As you can see I'm not very familiar with it.

[milahu]

umm, you should remove the commit 6491028
since the idx/sub files are protected by copyright
also the sub file has 5MB...

better use the test files from mpv
https://github.com/mpv-player/mpv/tree/master/test/samples

with strace i found this...

$ strace vobsub2srt patrol-mad.max1.1080p 2>&1 | grep -F .traineddata
openat(AT_FDCWD, "/nix/store/0zkrhb2g92d8yqk67vbh50mni1ax7gpj-tesseract-5.5.2/share/tessdata/deu.traineddata", O_RDONLY) = 3
^C

$ strace vobsub2srt --lang deu patrol-mad.max1.1080p 2>&1 | grep -F .traineddata 
openat(AT_FDCWD, "/nix/store/0zkrhb2g92d8yqk67vbh50mni1ax7gpj-tesseract-5.5.2/share/tessdata/div.traineddata", O_RDONLY) = 3
^C

$ strace vobsub2srt --lang de patrol-mad.max1.1080p 2>&1 | grep -F .traineddata 
openat(AT_FDCWD, "/nix/store/0zkrhb2g92d8yqk67vbh50mni1ax7gpj-tesseract-5.5.2/share/tessdata/deu.traineddata", O_RDONLY) = 3
^C

$ strace vobsub2srt --lang ger patrol-mad.max1.1080p 2>&1 | grep -F .traineddata 
openat(AT_FDCWD, "/nix/store/0zkrhb2g92d8yqk67vbh50mni1ax7gpj-tesseract-5.5.2/share/tessdata/eng.traineddata", O_RDONLY) = 3
^C

[csingleplus]

umm, you should remove the commit 6491028 since the idx/sub files are protected by copyright also the sub file has 5MB...

Done. Regarding your strace results, it's as I expected. I will need to add a routine that searches Nix stores for the Tesseract installation. Good find. I should be able to do that from my Linux install, but will test myself on Nix before I commit it.

[milahu]

Done

revert commit != remove commit

git branch bak-$(date -Is)
git reset --hard HEAD~2 # remove the last 2 commits
git push -f

I will need to add a routine that searches Nix stores for the Tesseract installation.

bad idea... the tessdata dir should be

• the default dir (which is found in my case, only the language selection is broken)
• the TESSDATA_PREFIX env
• the tesseract-data CLI parameter

so the only problem i see is the language selection

• lang=default → deu (good)
• lang=deu → div (bad)
• lang=de → deu (good)
• lang=ger → eng (bad)

[csingleplus]

I can confirm my build works with Nix. It just needs language detection fixed as you mentioned, which I'll look at next.