Remaining loaders/savers to rework for the source/target API #2051

kleisauke · 2020-01-30T21:07:36Z

kleisauke
Jan 30, 2020
Maintainer

This issue tracks the remaining loaders and savers that might be reworked for the new source/target API. I mostly used this comment as reference. Feel free to add more information in the comments.

Image suffix(es)	Loader	Saver	Finished
-	rawload	rawsave¹	Loader 🚧 Saver ✅
.v, .vips	vipsload²	vipssave²	Loader ⚠️ Saver ⚠️
whole-slide suffixes	openslideload	-	Loader ⚠️
.fits	fitsload²	fitssave²	Loader ⚠️ Saver 🚧
nifti suffixes	niftiload²	niftisave²	Loader ⚠️ Saver 🚧
IM suffixes	magickload (IM6) magickload (IM7)	magicksave	Loader 💡³ Saver 🚧

Loaders and savers missing from the above table: analyzeload, matload, openexrload.

Notes:

See PR Rework rawsave and add rawsave_target #3942.
We'd get early close for free if this is reworked.
The source is only loaded into memory if it's not file-based, see PR add magickload_source #4592.

Already supports the new source/target API

Image suffix(es)	Loader	Saver	Finished
.jpg, .jpeg, .jpe	jpegload	jpegsave	Loader ✅ Saver ✅
.png	pngload (libpng) pngload (libspng)	pngsave (libpng) pngsave (libspng)	Loader ✅ Saver ✅
.webp	webpload	webpsave	Loader 💡 Saver ✅
.gif	gifload⁵	gifsave⁶	Loader 💡 Saver ✅
.heif, .heic, .avif	heifload⁷	heifsave⁷	Loader ✅ Saver ✅
.jxl	jxlload	jxlsave	Loader ✅ Saver ✅
JP2K suffixes	jp2kload	jp2ksave	Loader ✅ Saver ✅
.pdf	pdfload (Poppler) pdfload (PDFium)	-	Loader ✅
.svg, .svgz, .svg.gz	svgload	-	Loader ✅
.csv	csvload	csvsave	Loader ✅ Saver ✅
.ppm, .pgm, .pbm, .pfm	ppmload	ppmsave	Loader 💡⁴ Saver ✅
.mat	matrixload⁸	matrixsave⁸	Loader ✅ Saver ✅
.hdr	radload	radsave	Loader ✅ Saver ✅
.tif, .tiff	tiffload	tiffsave⁹	Loader ✅ Saver ✅
.dz	-	dzsave¹⁰	Saver ✅

Notes:

The source is only loaded into memory if it can be efficiently memory-mapped and the PPM is binary (with a bit depth of at least 8 bits).
libnsgif needs the whole source file available at once, it can't stream images (as far as I know), so it's not a great fit. It's rather like webp. giflib isn't much better: you can only find out the number of frames by scanning the entire image, and you have to discover the number of frames as part of reading the header.
See upstream PR Add support for a user-supplied write callback function dloebl/cgif#7.
We could use the heif_context_read_from_reader and heif_context_write functions.
matrixload / matrixsave are used to read and write convolution matrices. They are very like the CSV load/save pair.
See PR Add tiffsave target #2798.
See PR add dzsave to a target #2718.

Legend:

Symbol	Meaning
✅	Source/target support available
💡	Needs to load the entire source into memory
⚠️	Only if there's a filename associated with the source/target
🚧	Source/target support not (yet) available

jcupitt · 2020-01-30T21:15:39Z

jcupitt
Jan 30, 2020
Maintainer

Oh hey well done! This is a great idea.

I've started revising gifload to work with streams:

https://github.com/libvips/libvips/tree/revise-gifload2

not done yet though.

0 replies

jcupitt · 2020-01-30T22:18:28Z

jcupitt
Jan 30, 2020
Maintainer

OK, gifload done and merged to master.

0 replies

kleisauke · 2020-02-05T14:42:32Z

kleisauke
Feb 5, 2020
Maintainer Author

Awesome! I confirm gifload_source is working well. Tested with:

wget https://storage.googleapis.com/downloads.webmproject.org/webp/images/dancing-banana.gif
cat dancing-banana.gif | vips copy stdin[n=-1] x.webp

I'll have a look at pdfload this week (will be quite similar to svgload, I think).

0 replies

jcupitt · 2020-02-21T18:24:07Z

jcupitt
Feb 21, 2020
Maintainer

I did csv load and save as well, and added csvload_source and csvsave_target.

0 replies

jcupitt · 2020-02-22T17:57:21Z

jcupitt
Feb 22, 2020
Maintainer

matrixload and matrixsave done too, plus _source versions.

0 replies

jcupitt · 2020-03-11T18:17:23Z

jcupitt
Mar 11, 2020
Maintainer

pdfload done, and I added pdfload_source.

0 replies

jcupitt · 2020-03-15T18:22:51Z

jcupitt
Mar 15, 2020
Maintainer

heifload and heifsave done too, plus _source/_target versions.

0 replies

kleisauke · 2020-06-25T19:11:25Z

kleisauke
Jun 25, 2020
Maintainer Author

I started to revise vipsload for the source API, see: kleisauke@a11b978.

While doing that, I noticed something strange in csvload_source and matrixload_source. I suspect that these lines:

libvips/libvips/foreign/matrixload.c

Line 405 in 1754087

if( matrix->source ) {

libvips/libvips/foreign/csvload.c

Line 603 in 1754087

if( csv->source ) {

Should check for source->source instead.

0 replies

jcupitt · 2020-06-25T21:54:49Z

jcupitt
Jun 25, 2020
Maintainer

Wow you're right, they are all messed up. Too much copy-paste programming :(

I've revised them. Thanks!

0 replies

jcupitt · 2020-06-25T22:04:23Z

jcupitt
Jun 25, 2020
Maintainer

... I'll try to add some tests too.

0 replies

jcupitt · 2020-06-27T14:28:01Z

jcupitt
Jun 27, 2020
Maintainer

OK, tests added, plus ppm load/save to/from a connection.

0 replies

kleisauke · 2020-12-30T15:29:42Z

kleisauke
Dec 30, 2020
Maintainer Author

I've added openslideload to the list and marked it as finished, since this is most likely done with commit babbb08.

If someone wants to work on revising vipsload, then see #1541 (comment). Otherwise, I'll pick this up in the course of next week.

0 replies

jcupitt · 2020-12-30T15:57:28Z

jcupitt
Dec 30, 2020
Maintainer

I'll add some tests for openslideload_source().

I'm reworking vipsdisp to use the new Source API, hence my current interest in this:

https://github.com/jcupitt/vipsdisp

0 replies

jcupitt · 2020-12-31T19:49:55Z

jcupitt
Dec 31, 2020
Maintainer

I did vipsload_source() and updated the table.

0 replies

jcupitt · 2021-01-02T15:53:09Z

jcupitt
Jan 2, 2021
Maintainer

... and vipssave_target.

0 replies

jcupitt · 2021-02-20T04:38:59Z

jcupitt
Feb 20, 2021
Maintainer

I did niftiload and fitsload. The savers still need doing.

0 replies

jcupitt · 2022-05-15T10:05:17Z

jcupitt
May 15, 2022
Maintainer

... tiffsave and dzsave now done.

6 replies

jcupitt May 15, 2022
Maintainer

The tiffsaver is a bit disappointing -- I hadn't realized that libtiff has to be able to read from the output. It'll write a chunk of data, write another chunk, then rewind and adjust some pointers. This means you have to buffer the whole output object in memory before you can start to write to a pipe :(

It's fine for filesystem output, of course.

kleisauke May 17, 2022
Maintainer Author

I think buffering the entire output object into memory is OK for non-file based targets. Perhaps we could use TIFFFdOpen() for file-based targets and TIFFClientOpen() for memory-based targets (similar to Pillow)? This would avoid having the ->read and ->seek in the target class, which is somewhat confusing.

jcupitt May 17, 2022
Maintainer

Yes, it's weird isn't it. I don't think anyone would design a format now that could only be written by something that could read its own output.

I wanted to make it work for subclasses too, so more than just memory and file. You only need to implement seek and read if you need tiff save support, so it's not too much of a burden, I hope.

kleisauke May 21, 2022
Maintainer Author

I wanted to make it work for subclasses too, so more than just memory and file. You only need to implement seek and read if you need tiff save support, so it's not too much of a burden, I hope.

Ah OK, seems reasonable to me. I had a quick attempt to implement this in NetVips and the weserv/images project, see commits:
kleisauke/net-vips@54a04c8
weserv/images@fabfb09

It works great! 🎉

jcupitt May 21, 2022
Maintainer

Great! It's in master now. I should test it in pyvips and ruby-vips too.

Uh oh!

Remaining loaders/savers to rework for the source/target API #2051

Uh oh!

Uh oh!

kleisauke Jan 30, 2020 Maintainer

Notes:

Notes:

Replies: 17 comments · 6 replies

Uh oh!

jcupitt Jan 30, 2020 Maintainer

Uh oh!

jcupitt Jan 30, 2020 Maintainer

Uh oh!

kleisauke Feb 5, 2020 Maintainer Author

Uh oh!

Uh oh!

jcupitt Feb 21, 2020 Maintainer

Uh oh!

jcupitt Feb 22, 2020 Maintainer

Uh oh!

jcupitt Mar 11, 2020 Maintainer

Uh oh!

jcupitt Mar 15, 2020 Maintainer

Uh oh!

kleisauke Jun 25, 2020 Maintainer Author

Uh oh!

jcupitt Jun 25, 2020 Maintainer

Uh oh!

jcupitt Jun 25, 2020 Maintainer

Uh oh!

jcupitt Jun 27, 2020 Maintainer

Uh oh!

kleisauke Dec 30, 2020 Maintainer Author

Uh oh!

jcupitt Dec 30, 2020 Maintainer

Uh oh!

jcupitt Dec 31, 2020 Maintainer

Uh oh!

jcupitt Jan 2, 2021 Maintainer

Uh oh!

jcupitt Feb 20, 2021 Maintainer

Uh oh!

jcupitt May 15, 2022 Maintainer

Uh oh!

jcupitt May 15, 2022 Maintainer

Uh oh!

kleisauke May 17, 2022 Maintainer Author

Uh oh!

jcupitt May 17, 2022 Maintainer

Uh oh!

kleisauke May 21, 2022 Maintainer Author

Uh oh!

jcupitt May 21, 2022 Maintainer

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

kleisauke
Jan 30, 2020
Maintainer

Replies: 17 comments 6 replies

jcupitt
Jan 30, 2020
Maintainer

jcupitt
Jan 30, 2020
Maintainer

kleisauke
Feb 5, 2020
Maintainer Author

jcupitt
Feb 21, 2020
Maintainer

jcupitt
Feb 22, 2020
Maintainer

jcupitt
Mar 11, 2020
Maintainer

jcupitt
Mar 15, 2020
Maintainer

kleisauke
Jun 25, 2020
Maintainer Author

jcupitt
Jun 25, 2020
Maintainer

jcupitt
Jun 25, 2020
Maintainer

jcupitt
Jun 27, 2020
Maintainer

kleisauke
Dec 30, 2020
Maintainer Author

jcupitt
Dec 30, 2020
Maintainer

jcupitt
Dec 31, 2020
Maintainer

jcupitt
Jan 2, 2021
Maintainer

jcupitt
Feb 20, 2021
Maintainer

jcupitt
May 15, 2022
Maintainer

jcupitt May 15, 2022
Maintainer

kleisauke May 17, 2022
Maintainer Author

jcupitt May 17, 2022
Maintainer

kleisauke May 21, 2022
Maintainer Author

jcupitt May 21, 2022
Maintainer