_Imported from Trac. This issue originally by _agoode* at 2010-04-28T14:36:16Z.*

The backends are relatively simple, they are just involved with assembling tiles into an image. A lot of work is involved in figuring out where the tiles actually need to go, and this is done in the vendor parsers. It might be interesting to add an option for the backends to dump out the tile positions as XML. Combined with dumping the properties and other metadata, this might be interesting.

_Imported from Trac. This issue originally by _agoode* at 2010-03-04T15:19:29Z.*

We should have man pages for the openslide- tools that are installed.

_Imported from Trac. This issue originally by _agoode* at 2010-04-28T14:30:44Z.*

It would be nice to be able to read_region into an allocated buffer that is larger than w_h_4. Since we are using cairo now, this should be extremely easy to do.

_Imported from Trac. This issue originally by _agoode* at 2008-11-06T22:49:59Z.*

Update the documentation in the distribution and the website to explain that we do not tell you about slide metadata at all, and we don't scale the image in any way.

_Imported from Trac. This issue originally by _agoode* at 2010-05-12T23:59:29Z.*

Right now, we load the associated images into memory at the beginning and hold onto them to memcpy them when asked. Now that we have error handling, we can load them on demand and report an error if that fails.

We should still save the dimensions at startup though, since the read_associated_image function does not take dimensions. This should be checked against what we read when called, and fail out if wrong.

_Imported from Trac. This issue originally by _agoode* at 2010-03-25T19:21:53Z.*

Win32 is not enough these days, we need win64 as well.

_Imported from Trac. This issue originally by _agoode* at 2010-05-11T14:20:33Z.*

The library is complicated enough, with enough error paths and formats that unit tests should be employed.

_Imported from Trac. This issue originally by _agoode* at 2009-12-11T20:04:16Z.*

Currently, we just layer over the tiles with overlaps in Trestle. But the vendor's software appears to do better.

_Imported from Trac. This issue originally by _anonymous* at 2010-04-26T20:01:24Z.*

test

_Imported from Trac. This issue originally by _agoode* at 2010-06-13T03:41:24Z.*

tilehelper will paint tiles from right to left, bottom to top. This seems to be needed for MIRAX + SATURATE. Is it really? Make sure this is clear in the code.

It seems that reading blocks backward could be a performance issue, especially on something like NGR format that is uncompressed on disk and read one scanline at a time.

_Imported from Trac. This issue originally by _agoode* at 2010-04-26T20:32:09Z.*

Hamamatsu VMS can contain files with distinct focal planes. We need another API to support this. We can avoid reading these other planes for now.

_Imported from Trac. This issue originally by _agoode* at 2009-07-17T18:39:44Z.*

_Imported from Trac. This issue originally by _agoode* at 2010-05-06T16:45:01Z.*

For fluorescence imagery, channels may not equal 3. If channels = 1, then we can use the existing read_region API to produce gray. Otherwise, some new API will need to be invented.

openslide_open may fail if channels != 1 or 3. We may need a new open function as well.

_Imported from Trac. This issue originally by _agoode* at 2010-07-14T14:21:04Z.*

From Hauke Heibel:

The inttypes guard (would also be required for stdbool which I removed
from openslide-jdatasrc.c) implies that we might consider providing a
single header file with these includes. Maybe openslide-private is the
one I am thinking about. Then it would be great if it were sufficient
for new modules to include this header file which should take care of
bools and inttypes and maybe also the third party includes. Thus we
could prevent new errors on different systems. Maybe one could also
consider having something like openslide-types that only deals with
the standard includes, bools and inttypes and maybe some macros.

_Imported from Trac. This issue originally by _agoode* at 2008-12-24T01:54:35Z.*

If we get a weird error, like when a file changes after opening, we should do something like cairo does and convert ourselves into an error object. Return 0 for width and height, "" for comment, and black for read_region.

_Imported from Trac. This issue originally by _agoode* at 2010-01-28T20:50:50Z.*

The openslide-testsuite contains anonymous pathology specimens useful for testing OpenSlide. It is large, but should be posted.

_Imported from Trac. This issue originally by _agoode* at 2009-07-17T18:38:35Z.*

Not necessary?

_Imported from Trac. This issue originally by _agoode* at 2010-06-01T21:01:33Z.*

Update the wiki to talk about our VMU support.

_Imported from Trac. This issue originally by _anonymous* at 2008-11-05T14:35:00Z.*

For some reason, an operation lead to corrupted files only in slide files I viewed with the wholeslide reader, HAMAMATSU format -- this may have occurred due to exiting the program mid-read during testing? I'm not sure why a file would change when "read"

_Imported from Trac. This issue originally by [email protected]* at 2009-05-20T03:14:40Z.*

There should be a mechanism to get misc. image meta data from the slide. OpenSlide Java binding API could for example return a Properties object. Meta data such as microns per pixel (MPP) should be exposed through API.

There is also should be a way to get slide label if that information is available. Aperio .svs images usually have a label embedded in the tiff image. Hammatsu images seem to point to some sort of slide map that contains it.

_Imported from Trac. This issue originally by _agoode* at 2010-05-04T17:16:02Z.*

The tiff backend will keep file descriptors open while the tiff object is allocated. This can be a limitation if many slides are read at the same time. I believe the jpeg backend does not do this.

_Imported from Trac. This issue originally by _agoode* at 2010-05-04T17:08:54Z.*

libtiff prints out lots of warnings without our knowledge. If libtiff provided a library- and reentrant-safe way of turning this off, we should do it.

_Imported from Trac. This issue originally by _agoode* at 2009-11-21T15:12:39Z.*

Add something like "openslide.guessed-mpp" for microns per pixel. I think we know how to do it for a few formats.

_Imported from Trac. This issue originally by _agoode* at 2009-07-16T18:36:09Z.*

Support MIRAX mrxs files.

_Imported from Trac. This issue originally by _agoode* at 2010-03-25T19:24:13Z.*

_Imported from Trac. This issue originally by _agoode* at 2010-06-21T15:08:29Z.*

There are a few issues with ndpi support. First, we need a custom TIFF parser, libtiff isn't up to the task, since it tries to validate things that we don't want validated. Next, we need a way to munge the JPEG stream to support JPEG files that are beyond the limit. I think once those things are taken care of, just using the existing JPEG backend will work.

_Imported from Trac. This issue originally by _agoode* at 2009-11-21T15:11:41Z.*

Put documentation of each format on the website.

In the CMU-1 dataset, the background is white. In OpenSlide, the destination buffer is set to black and this causes artifacts at the border of the scanmap since white and black pixels are blended resulting in grayish intensities.

As discussed, I tried to set the background color as long as it is defined by the file format. I tried to set the color using cairo functionality and my current implementation can be found at

SHA: hauke76@e69f2eda719294a10a88.

For some reason it does not work. Looking at the destination buffer after the call to cairo_paint, the colors are not set as I expected them to be. I followed the sample given here http://cairographics.org/FAQ/ under "How do I clear a surface?". I did not use source_rgb because I explicitly want to set the alpha value to zero and not to 1.0.

_Imported from Trac. This issue originally by _agoode* at 2010-06-01T16:44:24Z.*

I am not sure that TIFFRGBAImage always returns pre-multiplied RGBA data. Need to make sure this is always the case, or fix it up if not.

_Imported from Trac. This issue originally by _agoode* at 2009-07-17T18:37:26Z.*

If we host Win32 DLLs for openslide, we need the corresponding source for GPL/LGPL binaries.

_Imported from Trac. This issue originally by _agoode* at 2009-11-21T15:11:10Z.*

Trestle file, the layers don't fully match up, is this a problem with the slide file or the library?

_Imported from Trac. This issue originally by _agoode* at 2010-03-28T14:47:37Z.*

It looks like we might want to bundle up all the dependent DLLs that we will ship into an openslide SxS assembly. Is this a good idea?

It doesn't look like you can have private DLLs within an SxS assembly though, but will it all still work?

_Imported from Trac. This issue originally by _agoode* at 2010-01-27T21:39:31Z.*

Internally, we don't necessarily store filenames as absolute. This should probably be changed for robustness?

_Imported from Trac. This issue originally by _agoode* at 2010-05-04T17:11:15Z.*

It may be the case that we can expose some hints to the user via properties that allows greater performance when using openslide. These may include optimal width/height to read, x/y alignment, etc. This could correspond to tiles internally, though many formats have overlaps that would make this less useful.

_Imported from Trac. This issue originally by _agoode* at 2010-05-06T16:43:08Z.*

VMU format is 12-bit RGB. Truncating to 8-bit and using the existing read_region API is good enough for now, but we might want to give access to the higher-depth data in some way.

_Imported from Trac. This issue originally by _agoode* at 2009-07-17T18:36:12Z.*

Specfile is done, just need to get it reviewed.

_Imported from Trac. This issue originally by _agoode* at 2010-02-19T21:12:07Z.*

MIRAX shows seams on CMU-1.mrsx (and others) while displaying layer 1.

test

_Imported from Trac. This issue originally by _agoode* at 2010-03-31T03:58:09Z.*

We should fail early if there is a TIFF compression type that we cannot decode.

_Imported from Trac. This issue originally by _agoode* at 2010-05-04T17:13:42Z.*

libtiff does not allow multi-threaded reading of TIFF files. Either fix libtiff or replace it.

_Imported from Trac. This issue originally by _agoode* at 2010-05-04T17:12:56Z.*

Right now, we have a simple per-slide cache that is not shared. There should be a way to allow a user to do caching themselves, or not at all if it is a one-pass run. Perhaps the user can provide a cache themselves, run uncached and cache everything themselves.

_Imported from Trac. This issue originally by _agoode* at 2009-11-21T15:15:08Z.*

We don't always get the same x and y resolution, does this mean non-square pixels?

_Imported from Trac. This issue originally by _agoode* at 2010-06-09T16:44:23Z.*

The cairo branch contains a simple addition of openslide-cairo.h and openslide_cairo_read_region(). Consider merging this, but only after thinking about our assumptions of the image backend. (Mirax for one looks and prints horribly when drawn to PDF.)

_Imported from Trac. This issue originally by _agoode* at 2010-04-28T03:09:58Z.*

openslide-write-png would be a useful utility program that takes a file and writes out a section of it as a png.

_Imported from Trac. This issue originally by [email protected]* at 2008-11-06T12:10:28Z.*

The Hamamatsu format includes a .opt file, for "Optimisation". This lists offsets for some of the restart markers in the jpeg files.

Right now, we don't trust this file, since it only includes offsets for each row, not for each restart marker actually in the jpeg. And it seems the last offset is missing from the file.

Instead of scanning all the jpegs at startup (which is slow and very IO intensive), we should sparsely populate the internal offsets list with what we know is valid, and then fill in the knowledge as we discover more restart markers. Downsides are increased latency once the slide is open since we would now be reading less data up front.

_Imported from Trac. This issue originally by _anonymous* at 2008-11-06T19:54:25Z.*

_Imported from Trac. This issue originally by _agoode* at 2008-11-06T22:50:35Z.*

Make a little SDL viewer to demonstrate the C library.

_Imported from Trac. This issue originally by _agoode* at 2010-04-26T20:31:48Z.*

Hamamatsu VMS can contain files with distinct focal planes. We need another API to support this. We can avoid reading these other planes for now.

_Imported from Trac. This issue originally by _agoode* at 2008-11-20T14:32:15Z.*

ws_open is slow, there should be some mechanism for monitoring open progress. Or just make ws_open fast?

Original email:

The proposal:
+--------------[o][-][x]-+
  Loading Slide
  [XXXXXX   ] 75.2%
+-------------------------+
Please create an iterative loading process for the Wholeslide library that adds these two functions:

ws_open_setup()

ws_open_iterative( .., &float progress ) returns -1 when failed, 0 when
complete, ranking 0-1 floating point approximating progress through files based on lines/total filesize estimates

ws_open_end() if necessary, but might not be

This way I can create a process bar animation occurring during the initial
examination process by running this in a thread before triggering the Viewer widget.

_Imported from Trac. This issue originally by _agoode* at 2008-12-23T18:19:12Z.*

If attempting to read a JPEG that is bad, libjpeg will exit(). This can be avoided with use of setjmp/longjmp and overriding of the error structure. It is very important that the library not exit() unnecessarily.