Nice-to-have fix for autogen.sh

[Vitali Lovich]

On Feb 1, 2012, at 8:12 PM, Segher Boessenkool wrote:

>>>> Except for the fact that configure is really slow, I'd agree with you.
>>> 
>>> It doesn't take more than ten seconds on a really slow machine though,
>>> so it is almost bearable...
>> 
>> Actually, I'm well placed to know that configure is an order of magnitude slower (I'm not kidding) if you're using MinGW or cygwin on Windows.
> 
> On OSX it is ten times slower than on Linux already, but yes I remember
> builds on cygwin being ten times slower than on Linux (on Linux it was
> dominated by compiler time, on cygwin just by process startup).

> 
> I thought it would have improved a bit by now, and I thought mingw would
> be better.  I also thought the generally much faster hardware we have
> these days would have made it all sort-of a non-issue.  "Almost bearable".
configure is a shell script & thus fork heavy since it runs under bash - fork is ridiculously slow (i.e. if you rewrote the shell to spawn executables instead of forking you'd be faster).  So the story with mingw/cygwin will always be terrible with autoconf for the foreseeable future (& autoconf also does way more operations than are strictly necessary).

>> I could actually provide timings of the exact same machine (dual core, 4 GB RAM) running the same libusb configure if you want to see for yourselves.
> 
> Yeah I'd love to see some rough numbers.  autogen.sh (on a clean tree)
> and configure separately, both cygwin and mingw?
> 
>> So I will agree with Vitali that configure can be and is really slow on some platforms.
> 
> No one disagrees.  I wasn't aware of the _amount_ of pain you have to
> live with though!
It's somewhat exaggerated since you're typically not running configure all that frequently (although make is slow of course for the same reason), but yeah - it's quite annoying.  Even worse when you're under a VM.

-Vitali