texture2D gives wrong result in experimental backend shader interface

[nick87720z]

I tried to use texture() and then texture2D() for shader with experimental backend (at least it's better documented than legacy).
Originally I did noop shader based on example from manual:

vec4 window_shader() {
  return texelFetch( tex, ivec2( texcoord), 0);
}

That worked fine. Now I'm trying to do same with texture(), but it fills window with same color (though it's unique for entire window):

return texture( tex, texcoord);

I expected this to work in at least same way (no matter if default texcoord point to center or corner in the texel).
Trying to find hint in picom sources, I found this in shader.c:

gl_FragColor = vec4(texture2D(tex, vec2(texcoord.xy), 0).rgb, 1);

This snippet vec2(texcoord.xy) looks too redundant, at least for me it made no difference. Besides that my usage seems to match to this one even before I looked.
Is there need for more conversions to use texture() inside window_shader()?

[nick87720z]

Hm, discovered. Looks like texture() expects normalized coordinates format in range [0:1]. Passing divided by 100 coordinates gives some meaningful feedback (tile from window miniature).

Right usage:

return texture( tex, texcoord / textureSize( tex, 0));