Fix transparent surfaces occlusion (Issue #34)

[Paramveersingh-S]

This PR completely overhauls how order-independent transparency is handled in the rendering pipeline, specifically resolving Issue #34. Previously, a hack toggled the depth mask on a per-polygon basis inside the tessellator, which failed when opaque tracks were drawn before the transparent detector boundaries but failed the depth test.

To resolve this properly, this implements a Multi-Pass rendering strategy:

1. Pass 1 (Opaque): Draws only opaque elements (alpha == 1.0) with depth writing ON. Transparent elements are skipped. (Opaque objects like the yellow tracks render normally and populate the depth buffer).
2. Pass 2 (Transparent): Draws only transparent elements (alpha < 1.0) with depth writing OFF. Opaque elements are skipped. (Transparent elements like the blue/red detectors render over the yellow tracks without occluding each other).

Changes include:

• wsb.c: Refactored phg_wsb_traverse_all_postings into a two-pass architecture.
• wsgl.c & wsgl_attr.c: Added global wsgl_current_transparency_pass tracking and strict depth mask enforcement.
• wsgl_tess.c, wsgl_line.c, wsgl_marker.c, wsgl_text.c: Implemented pass-checking skip logic dynamically based on GL_CURRENT_COLOR or vertex colors.

@schwicke could you please compile and test this patch on your end with the scene from Issue #34 to verify the occlusion is resolved properly?

[schwicke]

Thanks once again for the contribution! This one still needs some iterations. If you run test_c1 (see attached output) you see that now all surfaces are rendered black. Lines do not turn up at all it seems, e.g. test_f1 stays entirely

What is remarkable is that first the scene seems to be rendered but then things get erased. This might be a buffer visibility issue, it looks as if the rendering buffer is visible, and then its swapped against something empty (?). This would explain at least what I see for the hourglass test (test_f4).

[Paramveersingh-S]

@schwicke Thank you for the test images.

The issue was exactly a buffer visibility problem caused by wsgl_begin_rendering and wsgl_end_rendering being called incorrectly inside the multipass loop. wsgl_begin_rendering forcefully clears both the color and depth buffers (via glClear), and wsgl_end_rendering swaps the buffers (via glXSwapBuffers).

Because my previous implementation wrapped each pass inside begin and end, what happened was:

1. Pass 1 (Opaque) drew the opaque elements perfectly and immediately swapped the buffers.
2. Pass 2 (Transparent) immediately started by clearing the buffer entirely (erasing Pass 1 completely) and then drew the transparent lines/polygons onto an empty black background, before swapping the buffer again.

I have pushed a new commit to this PR that fixes this by pulling wsgl_begin_rendering and wsgl_end_rendering outside of the multipass logic. To ensure the OpenGL state (matrix stack, colors, attributes) is still properly initialized for Pass 2 without clearing the buffers, I exposed the internal init_rendering_state function as wsgl_init_rendering_state in wsglP.h and called it between the passes.

This preserves the buffer between rendering passes while maintaining state consistency. Could you kindly compile this updated branch and re-run your test_c1 and test_f4 scenes?

[schwicke]

Hi,
Thanks for the fix. This works better, however, it does not solve the issue yet:

[schwicke]

Please, don't waste too much time with this one, it is not really important - just a nice to have.

Having said that, the latest patch comes with a new feature: somehow the clipping of the borders does not work as expected:

I have verified that this is specific to this patch set. The current main branch looks fine: