Details
Bug
Major
Description
I updated my 2.2.3 project to the 2.3M2. As soon as you make a clean and refresh of the dependencies, with the un-checked options every command from GGTS leads to the grails prompt presented. But the prompt can't be used to enter the commands, because this leads to error, that the build listener can't be load. This also happens with a newly created 2.3 project. Setting this setting back resolves the issue, but leads to many unneeded started Grails consoles . I even tested it with the latest nightly of IDE, Grails and Groovy, but all remains the same.
Activity
- All
- Comments
- Work Log
- History
- Activity
- Transitions
There's some special Windows escape code in GGTS. It apparantly there to avoid a bug in Java ProcessBuilder API where it would inadvertently split arguments that were meant to stick together.
The code essentially adds quotes around the command before sticking it in an argument to GrailsScriptRunner. It used to be that without these special handling ProcessBuilder on Windows would split the argument appart into multiple arguments.
Right now as I'm debugging however I can see that without the quotes process builder seems to actually do the right thing. Adding quotes no longer seems necessary and in fact causes a problem as it now passes the quotes to Grails which also treats them as escape sequences to keep the argument in one piece.
The bug in question is this one:
http://bugs.sun.com/view_bug.do?bug_id=6468220
I suspect now that the determining factor is the JVM version. I.e. need to determine what versions of JVM has the bug and only add the quoting as needed.
Kris De Volder
added a comment - - edited There's some special Windows escape code in GGTS. It apparantly there to avoid a bug in Java ProcessBuilder API where it would inadvertently split arguments that were meant to stick together.
The code essentially adds quotes around the command before sticking it in an argument to GrailsScriptRunner. It used to be that without these special handling ProcessBuilder on Windows would split the argument appart into multiple arguments.
Right now as I'm debugging however I can see that without the quotes process builder seems to actually do the right thing. Adding quotes no longer seems necessary and in fact causes a problem as it now passes the quotes to Grails which also treats them as escape sequences to keep the argument in one piece.
The bug in question is this one:
http://bugs.sun.com/view_bug.do?bug_id=6468220
I suspect now that the determining factor is the JVM version. I.e. need to determine what versions of JVM has the bug and only add the quoting as needed. Quick update.
I tried with different versions of JVM 1.7 and as far back as 1.6 u15. The behavior seems consistent.
I can fix the problem by disabling the special 'escape' code for windows. But as it was put in for a good reason way back I'm a bit reluctant still to turn it off without understanding potential repercussions.
I think the problem this escape code addressed was observed on Win XP. I will need to double check that removing the code doesn't create a problem for Win XP users (there may still be a few out there).
Kris De Volder
added a comment - Quick update.
I tried with different versions of JVM 1.7 and as far back as 1.6 u15. The behavior seems consistent.
I can fix the problem by disabling the special 'escape' code for windows. But as it was put in for a good reason way back I'm a bit reluctant still to turn it off without understanding potential repercussions.
I think the problem this escape code addressed was observed on Win XP. I will need to double check that removing the code doesn't create a problem for Win XP users (there may still be a few out there). Just tested on Win XP. Even on Win XP the special windows escaping code is not needed (and causes the bug when its there).
So I still don't know why exactly this workaround for http://bugs.sun.com/view_bug.do?bug_id=6468220
is no longer needed while it was at some time in the past.
I have verified that the ProcessBuilder bug still exists even JVM7!
With a breakpoint in ProcessBuilder I also found that the 'special quoting' are still being applied.
So presumably Eclipse has its own workaround for he processbuilder bug (GGTS doesn't use ProcessBuilder directly but via Eclipse).
If we are no longer exposed to this bug now but where in the past, maybe the workaround for the bug in Eclipse is only present
in some versions of Eclipse. That seems to be the only explanation left.
I'll need to check whether GGTS based on 4.2 and 3.8 are affected or not.
Kris De Volder
added a comment - Just tested on Win XP. Even on Win XP the special windows escaping code is not needed (and causes the bug when its there).
So I still don't know why exactly this workaround for http://bugs.sun.com/view_bug.do?bug_id=6468220
is no longer needed while it was at some time in the past.
I have verified that the ProcessBuilder bug still exists even JVM7!
With a breakpoint in ProcessBuilder I also found that the 'special quoting' are still being applied.
So presumably Eclipse has its own workaround for he processbuilder bug (GGTS doesn't use ProcessBuilder directly but via Eclipse).
If we are no longer exposed to this bug now but where in the past, maybe the workaround for the bug in Eclipse is only present
in some versions of Eclipse. That seems to be the only explanation left.
I'll need to check whether GGTS based on 4.2 and 3.8 are affected or not. Finally! Confirmation: this bug was fixed in Eclipse 4.3:
That explains why lots of people are hitting the problem with GGTS 3.3.0 RELEASE. It's the first release that offers a 4.3 based build.
It seems to properly close the GGTS bug here I need to deactivate my own ProcessBuilder bug workaround conditionally for Eclipse 4.3 but leave it active for 4.2, 3.8 (or older).
Kris De Volder
added a comment - Finally! Confirmation: this bug was fixed in Eclipse 4.3:
https://bugs.eclipse.org/bugs/show_bug.cgi?id=387504
That explains why lots of people are hitting the problem with GGTS 3.3.0 RELEASE. It's the first release that offers a 4.3 based build.
It seems to properly close the GGTS bug here I need to deactivate my own ProcessBuilder bug workaround conditionally for Eclipse 4.3 but leave it active for 4.2, 3.8 (or older). A proper fix is now committed. Closing issue.
Kris De Volder
added a comment - A proper fix is now committed. Closing issue. 
Kris - if you need me to test something for you on Windows, let me know.