Default height to VIPS_MAX_COORD for vips_thumbnail

[kleisauke]

This to prevent reduction in the vertical axis when the height is omitted. Marked this PR as draft since it changes the behavior of vips_thumbnail into a resize based on a specific axis (which is more common, I think) instead than a resize based on a square bounding box.

This PR resolves (partially) #709, but a better way to address that issue is to make width an optional parameter. This allows users to do this, for example:

// Resize to a width of 200 pixels, no matter what the height of the input image is
if( vips_thumbnail( filename, &image,
	"width", 200,
	NULL ) )
	return( -1 );

// Resize to a height of 200 pixels, no matter what the width of the input image is
if( vips_thumbnail( filename, &image,
	"height", 200, 
	NULL ) )
	return( -1 );

// Identity transform
if( vips_thumbnail( filename, &image, NULL ) )
	return( -1 );

Instead of the current behavior:

// Resize to a width of 200 pixels, no matter what the height of the input image is
if( vips_thumbnail( filename, &image, 200,
	"height", VIPS_MAX_COORD,
	NULL ) )
	return( -1 );

// Resize to a height of 200 pixels, no matter what the width of the input image is
if( vips_thumbnail( filename, &image, VIPS_MAX_COORD,
	"height", 200,
	NULL ) )
	return( -1 );

// Identity transform
if( vips_thumbnail( filename, &image, VIPS_MAX_COORD,
	"height", VIPS_MAX_COORD, 
	"size", VIPS_SIZE_DOWN,
	NULL ) )
	return( -1 );

This PR resolves kleisauke/net-vips#71 when merged.

[jcupitt]

I'm not sure, this seems like quite a large behaviour change :-( It's likely to break a lot of downstream code.

How about adding a new operation with the new behaviour? thumbnail_axis perhaps?

[kleisauke]

I'm also a little worried about this behavior/compatibility change. I created this PR mainly for reference purposes. I'll add the VIPS_MAX_COORD constant to NetVips instead.

A new thumbnail_axis function might be a little odd if it only sets the thumbnail->height here

libvips/libvips/resample/thumbnail.c

Lines 563 to 564 in 54bfa23

	if( !vips_object_argument_isset( object, "height" ) )
	thumbnail->height = thumbnail->width;

to VIPS_MAX_COORD instead. A new declaration which makes the width and height optional, for e.g.:

int
vips_thumbnail_axis( const char *filename, VipsImage **out, ... )

Is better, but having 2 functions that can do exactly the same thing is a bit weird.

Anyways, feel free to label this PR as snoozing. 😄

[jcupitt]

Yes, it would be odd :( Really we'd want to replace all the vips_thumbnail*() functions with your proposed new API, but we'd need to do it under a new name to avoid breakage.

vips_thumbnail2()? vips_load_and_resize()? They all seem ugly :(