Apologies for the poorly constructed patch email. My hope is to ban explicit #version strings entirely, because otherwise we'll end up with tests that needlessly break with certain APIs. The simplest way to avoid unnecessary #version strings appearing is to make shader_runner reject them all.
That said, I agree we want sanity tests to be able to specify a particular GLSL version. My original change introduced a new [require] element solely specifying the #version to be inserted, but we eventually settled on using the existing GLSL element and restricting the comparison operator to >=. The theory is that this simultaneously specifies the minimum GLSL implementation version required by the test and supplies the version to be inserted with the #version directive. If there is a case where these need to be different then this does break down and we need some new syntax, but the assumption was that they would always be the same. So I'm claiming the existing GLSL >= syntax covers the sanity test case and everything else too. I wasn't aware of the GL_ARB_ES..._compatibility extension testing requirements. That's useful to know. I concede that illegal/nonsensical combinations could arise from certain GL / GLSL version requirements, but I was assuming we would generally have sane .shader_test files. It's not as if shader_runner as it stands catches everything like that. On the trailing garbage issue, I actually had test code in there to check for this. I can put it back in. It looks as if the parser could be tightened up in various places. Stuart _______________________________________________ Piglit mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/piglit
