RFR: JDK-8211677: Java resource copy and clean should use MakeTargetDir macro (original) (raw)

Erik Joelsson erik.joelsson at oracle.com
Thu Oct 4 16:42:47 UTC 2018

Hello,

On 2018-10-04 04:47, Magnus Ihse Bursie wrote:

Looks good to me.

Have you searched the code for more instances of mkdir -p? Just did, lots of places. I will create a followup issue for this. The current issue is failing too many builds in our CI to not be fixed fast. Also, maybe we should check if using ACPROGMKDIRP would give us a better option on Solaris. (See https://www.gnu.org/savannah-checkouts/gnu/autoconf/manual/autoconf-2.69/htmlnode/Particular-Programs.html#Particular-Programs). It is supposed to give a non-race prone version of mkdir -p. Just to be clear: Even if we start using it for MKDIRP, I still think it's a good idea to use our macro. That's certainly interesting and should be investigated. I will add it to the followup issue.

/Erik

/Magnus

4 okt. 2018 kl. 00:51 skrev Erik Joelsson <erik.joelsson at oracle.com_ _<mailto:erik.joelsson at oracle.com>>:

Since upgrading our Solaris build machines to 11.3, we have experienced intermittent build failures in the recipes copying java resource files. It's unclear why this started happening now after having worked fine for so long, but it seems it's a race caused by concurrent calls to "mkdir -p". In other recipes we have worked around this by using a macro "MakeDir" which reduces the likelihood of concurrent calls happening. Rewriting these copy and clean rules to use the current preferred macros seem to alleviate the problem for us on Solaris 11.3, and also makes the build a little bit more coherent.

Bug: https://bugs.openjdk.java.net/browse/JDK-8211677 Webrev: http://cr.openjdk.java.net/~erikj/8211677/webrev.01/ <http://cr.openjdk.java.net/%7Eerikj/8211677/webrev.01/> /Erik

More information about the build-dev mailing list