Major ABI break in 7.1.1

[remicollet]

ImageMagick version

7.1.1-0

Operating system

Linux

Operating system, version and so on

Fedora

Description

In 6b2ae4e

- VERS_10.0 {
+ VERS_10.1 {

This is a major ABI break as linker consider everything as new symbols.

Notice: in RPM packaging, such thing is track

$rpm -q --provides ImageMagick-libs
libMagickWand-7.Q16HDRI.so.10(VERS_10.0)(64bit)
...

Also change in enums (pixel.h) is a ABI breakage, as value defined at buildtime may change at runtime

@@ -49,8 +49,8 @@
   IndexChannel = 0x0020,             /* Color Index Table? */
   ReadMaskChannel = 0x0040,          /* Pixel is Not Readable? */
   WriteMaskChannel = 0x0080,         /* Pixel is Write Protected? */
-  MetaChannel = 0x0100,              /* not used */
-  CompositeMaskChannel = 0x0200,     /* SVG mask */
+  CompositeMaskChannel = 0x0100,     /* SVG mask */
+  MetaChannel = 0x0200,              /* not used */
   CompositeChannels = 0x001F,
   AllChannels = 0x7ffffff,
   /*
@@ -89,8 +89,8 @@
   IndexPixelChannel = 5,
   ReadMaskPixelChannel = 6,
   WriteMaskPixelChannel = 7,
-  MetaPixelChannel = 8,
-  CompositeMaskPixelChannel = 9,
+  CompositeMaskPixelChannel = 8,
+  MetaPixelChannel = 9,
   MetaPixelChannels = 10,
   IntensityPixelChannel = MaxPixelChannels,  /* ???? */
   CompositePixelChannel = MaxPixelChannels,  /* ???? */

IMHO, such breaking changes have no value...

DrawInfo struct also have change (haven't dig a lot to see if this may break things, allocation seems done in the library, not in the user app)

Steps to Reproduce

$ objdump -T libMagickWand-7.Q16HDRI.so.10.0.0
...
00000000000a0420 g    DF .text  00000000000001a2  VERS_10.0   MagickWriteImageFile
000000000008e6d0 g    DF .text  000000000000012f  VERS_10.0   MagickEdgeImage
000000000009ea10 g    DF .text  0000000000000183  VERS_10.0   MagickSpliceImage
000000000007eda0 g    DF .text  000000000000b8fc  VERS_10.0   ImportImageCommand
...
$ objdump -T libMagickWand-7.Q16HDRI.so.10.0.1
...
00000000000a0420 g    DF .text  00000000000001a2  VERS_10.1   MagickWriteImageFile
000000000008e6d0 g    DF .text  000000000000012f  VERS_10.1   MagickEdgeImage
000000000009ea10 g    DF .text  0000000000000183  VERS_10.1   MagickSpliceImage
000000000007eda0 g    DF .text  000000000000b8fc  VERS_10.1   ImportImageCommand
...

Images

No response

[remicollet]

About map files, new entries may be added for new symbols (symbol versioning), but unchanged symbol have to stay in previous entry.

Of course ABI break are possible, but in such case, please bump all soname (MagickCore, MagickWand and Magick++)

[urban-warrior]

The enum change was required to properly support multi-spectral images. We anticipated some ABI breakage which we tried to account for by bumping the library revision. We did add a member to DrawInfo, but it was added as the last member and should not disrupt the ABI.

[urban-warrior]

To date we did not have proper support for an array of meta channels, e.g. meta channels 1-6. To add support we required that the meta channel be the last channel in the enum and additional meta-channels would follow, e.g., meta1, meta2, meta3, etc. Because of the ABI disruption, we think it is best to restore the previous enums so the ABI does not break and introduce a new enum to manage an array of meta channels. We'll commit your pull request, add a patch to refactor an array of meta channels, and issue a new release, hopefully within a day or two. Thanks for your assistance with this issue.