VIPS 8.17+ broke Lua bindings

[gcsideal]

When trying to run lua-vips testing, the output with VIPS 8.17+ is the following:

tests

Copying src/vips.lua src/vips/cdefs.lua src/vips/Connection.lua src/vips/gvalue.lua src/vips/Image.lua src/vips/Image_methods.lua src/vips/Interpolate.lua src/vips/log.lua src/vips/Source.lua src/vips/Target.lua src/vips/verror.lua src/vips/version.lua src/vips/vobject.lua src/vips/voperation.lua in vips/lua/lua-vips-1.1.12/jit-vips for test
****************** lua dynamic custom (jit) ******
++++++++++++++++++++++++++++++++++++++++++++++
(process:626950): VIPS-WARNING **: 11:14:55.614: Unexpected end of GIF source data
++++++++++++++++++++++++++++++++++++++++*+++++++++
95 successes / 0 failures / 1 error / 0 pending : 0.482333 seconds

Error -> spec/write_spec.lua @ 22
test image write to file can save and then load a png
./src/vips/voperation.lua:240: unable to call VipsForeignSaveSpngFile
vips_colourspace: no known route from 'matrix' to 'b-w'

Maybe I overlook something? Can this be fixed?

[rolandlo]

Thanks for reporting. I will investigate the issue.

[rolandlo]

I can reproduce the issue. Minimal reproducer (on Linux):

local vips = require "vips"
local im = vips.Image.new_from_array{ 1, 2, 3, 4 }
im:write_to_file("/tmp/x.png")

[kleisauke]

CLI reproducer:

$ vips black x.jpg 100 100
$ vips hough_line x.jpg x.png
vips_colourspace: no known route from 'matrix' to 'b-w'

I bisected this to commit libvips/libvips@9f590ce (/cc @jcupitt), but I'm not sure how this worked previously, since VIPS_INTERPRETATION_MATRIX is not something that a PNG image would typically store. For now, a possible fix might be:

--- a/spec/write_spec.lua
+++ b/spec/write_spec.lua
@@ -9,8 +9,7 @@ describe("test image write", function()
     end)
 
     describe("to file", function()
-        local array = { 1, 2, 3, 4 }
-        local im = vips.Image.new_from_array(array)
+        local im = vips.Image.black(100, 100)
         local tmp_png_filename = ffi.os == "Windows" and os.getenv("TMP") .. "\\x.png" or "/tmp/x.png"
         local tmp_jpg_filename = ffi.os == "Windows" and os.getenv("TMP") .. "\\x.jpg" or "/tmp/x.jpg"
 

(untested)

[gcsideal]

CLI reproducer:

$ vips black x.jpg 100 100
$ vips hough_line x.jpg x.png
vips_colourspace: no known route from 'matrix' to 'b-w'

Oh, so even the VIPS behavior has changed.
With version 8.16.1 your CLI test produced valid images, even if the PNG dimensions changed to 256x256 pixels.

[jcupitt]

This seems to fix it:

index 3a048586a..c5076bdee 100644
--- a/libvips/colour/colourspace.c
+++ b/libvips/colour/colourspace.c
@@ -439,7 +439,8 @@ static VipsColourRoute vips_colour_routes[] = {
  * vips_colourspace_issupported: (method)
  * @image: input image
  *
- * Test if @image is in a colourspace that [method@Image.colourspace] can process.
+ * Test if @image is in a colourspace that [method@Image.colourspace] can
+ * process.
  *
  * Returns: `TRUE` if @image is in a supported colourspace.
  */
@@ -449,13 +450,19 @@ vips_colourspace_issupported(const VipsImage *image)
        VipsInterpretation interpretation;
        int i;
 
+       interpretation = vips_image_guess_interpretation(image);
+
        /* Treat RGB as sRGB. If you want some other treatment,
         * you'll need to use the icc funcs.
         */
-       interpretation = vips_image_guess_interpretation(image);
        if (interpretation == VIPS_INTERPRETATION_RGB)
                interpretation = VIPS_INTERPRETATION_sRGB;
 
+       /* Treat MATRIX as B_W.
+        */
+       if (interpretation == VIPS_INTERPRETATION_MATRIX)
+               interpretation = VIPS_INTERPRETATION_B_W;
+
        for (i = 0; i < VIPS_NUMBER(vips_colour_routes); i++)
                if (vips_colour_routes[i].from == interpretation)
                        return TRUE;
@@ -516,6 +523,11 @@ vips_colourspace_build(VipsObject *object)
        if (interpretation == VIPS_INTERPRETATION_RGB)
                interpretation = VIPS_INTERPRETATION_sRGB;
 
+       /* Treat MATRIX as B_W.
+        */
+       if (interpretation == VIPS_INTERPRETATION_MATRIX)
+               interpretation = VIPS_INTERPRETATION_B_W;
+
        for (i = 0; i < VIPS_NUMBER(vips_colour_routes); i++)
                if (vips_colour_routes[i].from == interpretation &&
                        vips_colour_routes[i].to == colourspace->space)

ie. pretend MATRIX images are really B_W for colourspace conversions.

I'll make a libvips issue.

[gcsideal]

This seems to fix it:
[...]
ie. pretend MATRIX images are really B_W for colourspace conversions.

I've applied this on stock libvips 8.17.3 and can confirm it fixes this issue.

[jcupitt]

Oh drat, I was too sleepy and I pushed directly to 8.17 by mistake, I meant to make a PR. Anyway, there's a slightly different fix there now.

I credited you in the changelog, I hope that's OK.