The trouble is that a reasonable requirement such as single-precision floating point support would require a developer to copy/paste/modify hundreds of LOC. The

The trouble is that a reasonable requirement such as single-precision floating point support would require a developer to copy/paste/modify hundreds of LOC. The choice is really fork, duplicate or template. I don't blame glm for picking the latter. Every option is painful, but templates might be a little less so.