RFR: JDK-8189955 Configuration validation is broken for some types of paths

Erik Joelsson erik.joelsson at oracle.com
Thu Oct 26 08:09:18 UTC 2017

I see why this huge check becomes necessary for a valid comparison, but 
it's quite a bit of extra complexity added as well. I was worried about 
the extra overhead on an already slow platform so I took the time to 
measure. While it's certainly measurable it's hardly noticeable. On a 
fast machine, the difference is around 50-100ms on repeated make 
invocations (baseline 1.85s). On my slow laptop it's 300-400ms, but the 
baseline is already over 7 seconds so as said, hardly noticeable.

That said, line 392 is easily expressible using the make filter function 
instead of going to the shell. All the other shell calls seem to be 
necessary however.

Just thinking some more. Would it be possible to do something like "cd 
DIR && pwd" for both dirs and just compare the output of that instead? 
If that works it would seem like way fewer shell calls and simpler logic.


On 2017-10-25 17:03, Magnus Ihse Bursie wrote:
> We are validating that the topdir we're using when running make is the 
> same as when we were running configure.
> For some circumstances, most typically when using a subdirectory in 
> the user's home directory (e.g. ~/jdk10) on Cygwin, this check fails.
> Bug: https://bugs.openjdk.java.net/browse/JDK-8189955
> WebRev: 
> http://cr.openjdk.java.net/~ihse/JDK-8189955-configuration-validation-is-broken-for-some-paths/webrev.01
> /Magnus

More information about the build-dev mailing list