On Jan 28 2014, at 17:50 , Magnus Ihse Bursie <magnus.ihse.bur...@oracle.com> wrote:
> > On 2014-01-29 01:44, Mike Duigou wrote: >> Looks fine. (I have been caught by the Windows cygwin issue as well). >> >> Do you want to do away with the performance hints entirely? I had another >> patch for setting memsize which added other hints. > Yes, I thought we should remove it entirely. I've heard people complain about > the .hide-performance-hints file it creates, and frankly, there's probably > too much information already in the configure output -- I don't think anyone > really reads it. Fredrik put up the hint long time ago, before we started > seeing all these issues with ccache, and I was not courageous enogh to remove > it earlier. :-) > > But if you have another patch that relies on it, I'll either leave it in > place, or you'll have to resurrect it later. Remove it. I didn't like the aspect of hiding the hints on subsequent configure runs. > Maybe it would be better to have a separate way to answer the question "how > can I increase build performance"? Maybe some kind of tool that analyses > common performance issues with the machine and/or the configuration? Rather > than to put a lot of output from configure, where even the important stuff > drowns... We already spew a lot during configure. I don't think we should be shy about including warnings if they are relevant. People can just ignore them if they wish (at the price of certain peril). Mike > /Magnus > >> >> Mike >> >> On Jan 28 2014, at 15:31 , Magnus Ihse Bursie >> <magnus.ihse.bur...@oracle.com> wrote: >> >>> Bug: https://bugs.openjdk.java.net/browse/JDK-8027584 >>> >>> While ccache can be used to speed up recompilations, there are also several >>> issues with ccache. Old versions does not work correctly for the jdk (we >>> try to detect that). It does not work well on other platforms than Linux; >>> on Windows it does not work at all. Nevertheless is ccache enabled on >>> Windows if it is installed, which it apparently tends to be with Cygwin >>> nowadays, requiring the use of --disable-ccache to avoid compilation >>> failure. >>> >>> In many circumstances ccache also does not provide any performance benefit. >>> For instance, new compilations gets a performance hit. Only recompilations >>> benefits. >>> >>> It is better to turn this upside down. This patch makes ccache disabled by >>> default, on all platforms. If you want ccache and are sure you are in a >>> situation that benefits from it, then you can enable it. It also removes >>> the output in configure hinting that ccache should be used. >>> >>> WebRev: >>> http://cr.openjdk.java.net/~ihse/JDK-8027584-disable-ccache-by-default/webrev.01 >>> >>> /Magnus >
