Re: Are "incomplete" headers bugworthy?
On Tue, 01 Apr 2003 17:47:39 +0000, Steve Greenland wrote:
> While I prefer to do as you suggest in my code, there is a significant
> faction that disagrees, and a substantial body of code that would be
> affected. You might check out the last section of
> http://www.lysator.liu.se/c/pikestyle.html for the reasoning.
... which is from 1989. At that time, lexical analysis may well have been
the most expensive part of compilation, but these days, there are more
important concerns. Optimization, for example. Or programmers' brain
capacity. (Besides, gcc does NOT re-read #ifndef-FOO-protected include
Why the hell should I update _all_ programs which #include <foo.h> just
because the author of foo.h changed something which now requires <bar.h>?
(Such changes don't necessarily affect the API/ABI.) Maybe Rob Pike's Unix
system didn't change in any significant way during the ten years before
he wrote that essay, but mine sure did during the ten-plus years afterwards.
> The solution to all autoconf problems is to remove autoconf and write
> properly portable code and build systems. But that also is an upstream
That would be nice. I think I'll just [ ask upstream to ] remove the
offending test from configure.in, or something...