Improve shader logging

[VReaperV]

Parse info logs on shader compilation failure in an implementation-dependent way, which poduces an output like this:

 136: 	vec4 color;
 137: 	vec2 texCoord, lmCoord;a
 138: 
 139: 	VertexFetch( position, LB, color, texCoord, lmCoord );
------	-----------^-------------------------------------------------------
0(139) : error C0000: syntax error, unexpected '(', expecting "::" at token "("

 140: 	color = color * u_ColorModulate + u_Color;

If the exact character is not known, the line will just be like "^^^^^^^^^^".

Also reorganised some code a bit due to changes. Removed BuildGPUShaderText() since it was just unnecesserily fragmenting the code at this point.

[VReaperV]

I have tested this on Nvidia and Intel (the latter sometimes reports incorrect line numbers though...), so just need someone to test with AMD.

[illwieckz]

Mesa is probably all the same whatever the hardware (AMD, Nvidia, Intel, etc.), and Mesa is probably different different to Windows/macOS proprietary AMD even on AMD. With some luck AMD OGLP on Linux is same as AMD on Windows and macOS. On Windows and Linux, one may try Mesa llvmpipe as a generic Mesa driver.

[slipher]

Guess I'll wait for this to be rebased on #1242 :)

[VReaperV]

Rebased. This now uses the functionality added in #1242.

[slipher]

Neat, works for me on Nvidia and Intel. Well, sort of works on Intel, as promised :)

[slipher]

Why is a text with 0 newlines a special case?

[VReaperV]

They either added a newline if empty or affected which line was reported by the driver, I'll add a comment for whichever it was.

[VReaperV]

So this is actually because each \n corresponds to 2 lines so any header with at least 1 needs to have its line count increased by 1.

[slipher]

The driver can't know whether the input had an empty string concatenated in it. Maybe it's just the "compat header" adding one by being empty, canceling out the -1 subtracted from all the line count sums.

[VReaperV]

So I checked everything again, and it turns out that what was throwing me off were the shader compile macros. They're never passed to InitShader(), but they are added when the shader code is actually sent to the driver to compile. This is a problem on master as well w. r. t. #insert getting incorrect line numbers because of this, so I've removed the erroneus line calculation, but the bug resulting from macros is still there.

[VReaperV]

I've opened an issue for it: #1339. I consider that out of scope here though, since this pr is about logging.

[slipher]

The driver can't know whether the input had an empty string concatenated in it. Maybe it's just the "compat header" adding one by being empty, canceling out the -1 subtracted from all the line count sums.

[slipher]

LGTM