ARB Meeting Notes
March 9-10, 1998
Hosted by Evans and Sutherland in Salt Lake City, UT
Meeting notes taken by Jon Leech, SGI
|Bill Armstrong||Evans & Sutherlandfirstname.lastname@example.org|
|Craig Groeschel||Metro Linkemail@example.com|
|David Ligon||Raycer Graphicsfirstname.lastname@example.org|
|Fred Fisher||Dynamic Picturesemail@example.com|
|Garry Paxinos||Metro Linkfirstname.lastname@example.org|
|Nate Robins||University of Utahemail@example.com|
Kurt Akeley from SGI summarized Fahrenheit. There are three components:
SGI felt it was critical to work together with Microsoft, which had not been moving in compatible directions, to be able to build value-added products in the Windows environment. Fahrenheit is a large step in that direction.
The ARB has made several attempts to get agreement on high-level APIs, and has overall failed, sometimes in a frustrating manner. SGI's experience is that apps need to write to a scene graph to get good performance without micro-optimization. But it's also important to be able to "reach around" the scene graph to the low level. SGI's motivation with Fahrenheit was primarily to get agreement on a good scene graph.
What happens with OpenGL and OpenGL++?
Steve Wright from Microsoft followed up:
This was the major activity of the meeting, taking most of the day and being continued Tuesday morning. Rather than taking detailed minutes of the discussions, a separate summary of the changes to be applied to the final pre-meeting draft of the spec is provided.
DEC was unable to attend this meeting; they were brought in at times for conslution and votes by teleconference. This was especially important since most of the major issues remaining with the specification were raised by Andy Vesper at DEC.
One significant item had to be decided by ARB vote: the set of convolution border modes to be included in OpenGL 1.2, and the exact definition of those modes. The three votes were:
David Yu led this discussion. GLX 1.3 still has a number of open issues to be resolved, which are summarized in the GLX minutes.
The GLX protocol document has not yet been updated based on the protocol for the OpenGL 1.1 extensions on which 1.2 features were based.
GLX discussion will be continued on the mailing list.
Bill Armstrong organized dinner at the Old Salt City Jail.
Bill Mitchell's FORTRAN 90 bindings were approved by the ARB in February. The specification is kept on opengl.org, and his sample implementation is available from http://math.nist.gov/f90gl/.
As many people know by now, GLUT author Mark Kilgard is no longer working at SGI. This has led to some uncertainty regarding the status of GLUT. SGI stated its official position:
SGI believes that GLUT is a powerful tool for OpenGL programming, and wants to see it continue to evolve in the future. However, SGI does not own GLUT; Mark Kilgard holds the copyright and will determine what happens with GLUT through his actions.
SGI will continue to provide space for Mark's GLUT pages (http://reality.sgi.com/mjk/glut3/glut3.html) on its employee personal web server, as a service to the OpenGL community. Mark says that he's still working on a new GLUT revision (GLUT 3.7), despite his change in employment. When future GLUT versions appear, they will also be hosted on reality, if Mark so wishes.
SGI expects to update the Red and Blue books for OpenGL 1.2, though details have not been finalized. The Green book (OpenGL Programming for the X Window System) was done by Mark Kilgard; we'll check with him to see if he plans to update it. Likewise for Ron Fosner's White book (OpenGL Programming for Windows 95 and Windows NT).
SGI is aiming at simultaneous release of 1.2 sample GL, conformance tests, and GLX 1.3 towards end of summer '98. Conformance proposals and the GLX 1.3 specification and protocol document must be completed in the near future, since they are required to finish the sample GL. Aid from other ARB members and licensees is welcome, particularly in finalizing conformance proposals and writing code for them.
IBM noted that the sample GL sets the internal format of the proxy texture to 0 when the proxy query fails, while IBM's behavior is to set the internal format to the specified format. Do we want to change the sample? Tabled for now.
DEC was brought in by teleconference for most of this discussion. After deciding on a few final points, a vote for approval of OpenGL was held. The vote was on the March 4 draft together with the list of changes resulting from the discussion.
The vote was unanimous in favor; OpenGL 1.2 is approved. The final specification document will be posted by Jon Leech shortly after the meeting.
Both the general strategy of conformance testing and specific proposals for testing OpenGL 1.2 features were discussed.
David Yu & Dan Brokenshire discussed the GLX conformance shell, in particular how to test all potentially possible FBConfigs. This is complicated and will be taken offline.
Dan Brokenshire expressed some willingness by IBM to share the load of developing conformance test code (as well as specs).
Dave Ligon asked about OPC posting of conformance tests. We've previously established that OPC is not willing to do this. Another possibility, posting on opengl.org, raises a conflict of interest issue - opengl.org is funded by SGI. However, SGI is willing to set up links from opengl.org to vendor-hosted conformance results.
There was limited interest (5 people) in public posting of conformance results. IHVs can get Windows conformance binaries and sources through Microsoft. Unix vendors can choose or not to publish their results.
Evans & Sutherland raised the issue of minor problems with conformance. The big issue appears to be the lag between submitting bugs to the ARB and the resulting updated conformance suite passing through SGI to Microsoft and WHQL. People should discuss this with Steve Wright if the lag in fixes is a problem.
Suggested changes (diffs, included files) to conformance tests should be sent to firstname.lastname@example.org only, not email@example.com. But it's not a criminal act if an accident happens.
The mustpass tests are really aimed at making sure that implementations don't omit features, while the 'regular' conformance tests establish accuracy. We need to discuss this on opengl-arb-interest after the meeting, to establish a clear demarcation of mustpass from conformance tests. Jon Leech will push this after the ARB meeting.
All tests should add their enumerants and functions to a 1.2-specific coverage test. Imaging features will be tested only if the implementation supports the imaging subset (as defined by the GL_EXTENSIONS string).
|BGRA Pixel Formats (SGI)||bgra_test.txt|
|Convolution Border Modes
(Sun & HP)
|Draw Range Elements (SGI)||drawrange_test.txt||Might this go into primtest? There's no existing test for DrawElements().|
|Packed Pixel Formats (Nvidia)||packed_test.txt||The problem is the large number of formats supported.|
|Rescale Normal (IBM)||rescale_test.txt||The proposal suggests modifying the xformn test, which we probably don't want to do. Create a parallel test specific to rescale_normal instead?|
|Separate Specular Color (E & S)||specular_test.txt||E & S feels that mustpass should test minimal functionality, not comprehensive. Bill is reluctant to put in new color clipping requirements on the secondary color that aren't there for the primary color. He'll look more closely at mustpass. Bimal thinks there are no color clipping tests in the 1.1 suite. However, IBM & Intel both feel that simple clipping tests would be appropriate.|
|Texture Edge Clamp (SGI)||To Be Done|
|Texture LOD (SGI)||To Be Done|
|Texture 3D (Real 3D)||texture3D_test.txt||As a PC vendor driven by WHQL testing, Real 3D was not totally clear on the distinction between mustpass and "GLLite" tests, and how this maps onto mustpass / conformance tests.|
|Imaging Subset Features|
Blend Equation (SGI)
|Color Matrix (SGI)||colormatrix_test.txt|
Color Subtable (SGI)
|colortable_test.txt||Should be split into mustpass and conformance components.|
|Convolution (SGI)||convolution_test.txt||Should be split into mustpass and conformance components - there are a total of 9 tests.|
|To Be Done|
|FB Config (SGI)||To Be Done|
|Make Current Read (SGI)||To Be Done|
|Pbuffer (SGI)||To Be Done|
The idea of ARB extensions is to have a new level of formalism and support beyond EXT. They will have a specification agreed on by the ARB, conformance tests, and a sample implementation provided with the sample GL. They do not have to be supported by licensees, and they are not part of the OpenGL specification itself.
Extensions discussed were:
Multitexture will be pursued by email and otherwise offline by interested IHVs and ISVs. Nvidia will host a first meeting in Sunnyvale next week,
Sun is proceeding with AWT bindings for JDK as discussed in the December meeting. The bindings are to appear in JDK 1.2 beta 3. While Sun can't preannounce dates, the beta will probably ship in the next couple of months, and final will probably be available towards the end of summer '98.
Arcane Technologies is interested in submitting Magician, their Java bindings, to the ARB as a standard. They prepared a document describing the Magician interface which was distributed at the meeting, and will attend the June ARB meeting to discuss the proposal. In the meantime, the ARB will discuss Magician with Arcane via the opengl-arb-interest list, to learn more about it and about the conditions under which it may be made available to licensees and the ARB.