Community
Participate
Working Groups
I'm not sure how or when our logs file has gotten so large, but for a full build, the main log file (antBuilderOutput.log) is about 130M. Not that much smaller than the code distribution! These used to be about 30M. Not sure exactly what's changed ... perhaps some behavior of pde build or ant? But, at at rate, we should make efforts to make much smaller and/or be able to better control what's logged.
One thing that helps (at least then tailing log) is to use a little-documented flag to pde build cvs.reallyquiet=true see bug 256379
Another thing that has helped is to use -quiet directive on ant invocations. We had been doing this where we use ant command directly, but not when we invoke antRunner eclipse application. I've "tied" this use of quiet to a variable we were already using "USE_QUIET" so it can be turned on and off. With if off, the logs go down to about 90M.
There is a lot of unnecessary "echo" statements, at the "top" of many ant files (outside of a target). These print too much info, over and over again. We could put these in "init" methods (or similar) and call only once, setting variables where needed, and then setting an "hasBeenIntialized" variable, so is not executed repeatedly.
There also appears to be huge savings if we stop having the compiler produce "discouraged access" warnings ... see bug 339817. Some local experiments makes it appear these warnings, for a full build, take about 80 M of logging space.
Just to document it, here's the "history" of the log size, looking at our releases in 'archives'. Notice the huge jump in the last release, R3.2.3. If necessary, or of sufficient interest, I suspect we could do some compares and figure out what changed. But, even 30M is bad enough, so should reduce/control in a much better way. 1M ./R0.5/R-0.5-20080820002020/antBuilderOutput.log 20M ./R2.0/R-2.0.2RC3-20080221101112/antBuilderOutput.log 20M ./R2.0/R-2.0.2-20080223205547/antBuilderOutput.log 3M ./R2.1/R-2.1.1-20090226062131/antBuilderOutput.log 3M ./R2.1/R-2.1-20081218190237/antBuilderOutput.log 30M ./R3.0/R-3.0-20080616152118/antBuilderOutput.log 26M ./R3.0/R-3.0.5-20090521045405/antBuilderOutput.log 26M ./R3.0/R-3.0.4-20090213193639/antBuilderOutput.log 97K ./R3.0/R-3.0.2-20080921203356/antBuilderOutput.log 26M ./R3.0/R-3.0.1-20080818032401/antBuilderOutput.log 26M ./R3.0/R-3.0.3-20081113203138/antBuilderOutput.log 30M ./R3.1/R-3.1.1-20090917225226/antBuilderOutput.log 29M ./R3.1/R-3.1-20090616035105/antBuilderOutput.log 29M ./R3.1.2/R-3.1.2-20100211202452/antBuilderOutput.log 33M ./R3.2.0/R-3.2.0-20100615235519/antBuilderOutput.log 32M ./R3.2.1/R-3.2.1-20100730021206/antBuilderOutput.log 32M ./R3.2.2/R-3.2.2-20100915173744/antBuilderOutput.log 105M ./R3.2.3/R-3.2.3-20110217214612/antBuilderOutput.log
Status: Huge help was to no longer capture or log "discouraged access" warnings. Easy to turn off, by using -discouraged. Still need to fix our custom ant task that parses compile logs to produce summary table. Currently shows up as 0 discouraged access warnings, which is kind of misleading. Should just leave out of table. I've consolidated a lot of "echos" so they are either done once (per build process) or only if in debug mode. One thing, just noticed, is when we sign we "updatePackProperties", part of which is to unzip the zip to be signed, to look for jars/bundles that are already signed. By having verbose="true" causes each file to be listed, which adds up to about 12 Megs of log space for WTP ... so will use "verbose="false" in builds staring next week.
I'll count this as fixed, reduced as much as practical for now. The access/discouraged access issue/improvements will continue to be tracked in bug 339817.
