On 5 March 2012 23:16, Eric Anholt <<a href="mailto:eric@anholt.net">eric@anholt.net</a>> wrote: <div class="gmail_quote"><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> <div class="im">On Fri, 2 Mar 2012 15:40:28 -0800, Paul Berry <<a href="mailto:stereotype441@gmail.com">stereotype441@gmail.com</a>> wrote: > This is the first in a series of patches that replaces GLEW with a new > mechanism called "piglit-dispatch". This patch adds the script that > will be used by piglit-dispatch to build its generated code. > > The input to the code generation script is a set of XML files > describing the GL API, in the same form as used by Mesa. For the time > being, these files can be found at > git://<a href="http://github.com/stereotype441/glapi.git" target="_blank">github.com/stereotype441/glapi.git</a>. </div>From talking today, it sounds like investigating glspec as a source for this instead of the XML is very promising. But I'll review it anyway. I have to say: This was really nice code to read. I'm usually not a fan of python (I tend to find python code impenetrable), but I feel like I understand what this code is doing, and I understand python in general better as a result. </blockquote><div> Thanks :) </div><blockquote class="gmail_quote" style="margin:0pt 0pt 0pt 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"> <div class="im"> > +# Generate a top-of-file comment cautioning that the file is > +# auto-generated and should not be manually edited. > +def generated_boilerplate(): > + return """\ > +/** > + * This file was automatically generated by the script {0!r}. > + * > + * DO NOT EDIT! > + * > + * To regenerate, run "make piglit_dispatch_gen" from the toplevel directory. > + */ > +""".format(os.path.basename(sys.argv[0])) </div>Even though the files are generated, and they're describing an API in a way that may or may not be copyrightable, it's probably a good idea to put a license on it so people know they can at least use it in the usual permissive way. </blockquote><div> Yeah, that's a good point. I'll put the usual Intel copyright notice in. </div><blockquote class="gmail_quote" style="margin:0pt 0pt 0pt 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"> <div class="im"> > + # Generate a list of Function objects representing all functions > + # in the API. The resulting list is sorted by function name. > + def compute_unique_functions(self): > + functions = [] > + names_used = set() > + for f in self.__functions: > + if <a href="http://f.name" target="_blank">f.name</a> in names_used: > + continue > + functions.append(f) > + functions.sort(key = lambda f: <a href="http://f.name" target="_blank">f.name</a>) > + return functions </div>I was surprised to see this function -- I would have assumed that the parsed Api contained uniquely-named Functions in the first place. Is it to do with parsing GL/GLES variants of XML or something? Are those all exact matches in terms of signature? </blockquote></div> When I wrote this I wasn't sure whether there were uniquely-named functions in the XML or not. It turns out that there currently aren't. But I expect we'll need to add a few. For example, the BindBufferBase() function is defined in OpenGL 3.0 and in ARB_uniform_buffer_object, and it's allowed for an implementation to provide either OpenGL 3.0, ARB_uniform_buffer_object¸ or both. That means that in order for piglit-dispatch to handle all three possibilities correctly, we'll have to specify the XML for BindBufferBase() in both the GL 3.0 and ARB_uniform_buffer_object categories. I think there are other examples like this too, but I haven't done a comprehensive search.