Re: svn commit: r645722 - /tomcat/trunk/java/org/apache/catalina/connector/InputBuffer.java
just taking a quick look at the patch, this is a memory leak.
the buffer will simply grow larger and larger until an OOM happens, if
one keeps marking the buffer
when I am more awake, I'll help out on this, it's very easy to reproduce
http://people.apache.org/~fhanik/upload-mark.jsp
Filip
[EMAIL PROTECTED] wrote:
Author: markt
Date: Mon Apr 7 15:47:54 2008
New Revision: 645722
URL: http://svn.apache.org/viewvc?rev=645722view=rev
Log:
All of the issues I was seeing with mark/reset were due to states that resulted
in a call to CharChunk.flushBuffer().
I tried many different ways to fix it but by far the simplest was this patch
that just increases the size of the CharChunk internal buffer when creating the
mark sufficiently that flushBuffer() is never called.
With this patch I can't break mark/reset with bug 44494's test case in single
or multibyte mode.
Modified:
tomcat/trunk/java/org/apache/catalina/connector/InputBuffer.java
Modified: tomcat/trunk/java/org/apache/catalina/connector/InputBuffer.java
URL:
http://svn.apache.org/viewvc/tomcat/trunk/java/org/apache/catalina/connector/InputBuffer.java?rev=645722r1=645721r2=645722view=diff
==
--- tomcat/trunk/java/org/apache/catalina/connector/InputBuffer.java (original)
+++ tomcat/trunk/java/org/apache/catalina/connector/InputBuffer.java Mon Apr 7
15:47:54 2008
@@ -457,11 +457,7 @@
cb.setOffset(0);
}
}
-int offset = readAheadLimit;
-if (offset size) {
-offset = size;
-}
-cb.setLimit(cb.getStart() + offset);
+cb.setLimit(cb.getStart() + readAheadLimit + size);
markPos = cb.getStart();
}
-
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
-
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
Re: svn commit: r645722 - /tomcat/trunk/java/org/apache/catalina/connector/InputBuffer.java
ignore my prev email about the mem leak, I wasn't up to current in
trunk, and was running into something else
Filip
[EMAIL PROTECTED] wrote:
Author: markt
Date: Mon Apr 7 15:47:54 2008
New Revision: 645722
URL: http://svn.apache.org/viewvc?rev=645722view=rev
Log:
All of the issues I was seeing with mark/reset were due to states that resulted
in a call to CharChunk.flushBuffer().
I tried many different ways to fix it but by far the simplest was this patch
that just increases the size of the CharChunk internal buffer when creating the
mark sufficiently that flushBuffer() is never called.
With this patch I can't break mark/reset with bug 44494's test case in single
or multibyte mode.
Modified:
tomcat/trunk/java/org/apache/catalina/connector/InputBuffer.java
Modified: tomcat/trunk/java/org/apache/catalina/connector/InputBuffer.java
URL:
http://svn.apache.org/viewvc/tomcat/trunk/java/org/apache/catalina/connector/InputBuffer.java?rev=645722r1=645721r2=645722view=diff
==
--- tomcat/trunk/java/org/apache/catalina/connector/InputBuffer.java (original)
+++ tomcat/trunk/java/org/apache/catalina/connector/InputBuffer.java Mon Apr 7
15:47:54 2008
@@ -457,11 +457,7 @@
cb.setOffset(0);
}
}
-int offset = readAheadLimit;
-if (offset size) {
-offset = size;
-}
-cb.setLimit(cb.getStart() + offset);
+cb.setLimit(cb.getStart() + readAheadLimit + size);
markPos = cb.getStart();
}
-
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
-
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
Re: svn commit: r645722 - /tomcat/trunk/java/org/apache/catalina/connector/InputBuffer.java
On Mon, 2008-04-07 at 22:47 +, [EMAIL PROTECTED] wrote: Author: markt Date: Mon Apr 7 15:47:54 2008 New Revision: 645722 URL: http://svn.apache.org/viewvc?rev=645722view=rev Log: All of the issues I was seeing with mark/reset were due to states that resulted in a call to CharChunk.flushBuffer(). I tried many different ways to fix it but by far the simplest was this patch that just increases the size of the CharChunk internal buffer when creating the mark sufficiently that flushBuffer() is never called. With this patch I can't break mark/reset with bug 44494's test case in single or multibyte mode. Oops, I had forgotten that part, there was indeed another problem with that. Sorry. Rémy - To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
Re: svn commit: r645722 - /tomcat/trunk/java/org/apache/catalina/connector/InputBuffer.java
Remy Maucherat wrote: On Mon, 2008-04-07 at 22:47 +, [EMAIL PROTECTED] wrote: Author: markt Date: Mon Apr 7 15:47:54 2008 New Revision: 645722 URL: http://svn.apache.org/viewvc?rev=645722view=rev Log: All of the issues I was seeing with mark/reset were due to states that resulted in a call to CharChunk.flushBuffer(). I tried many different ways to fix it but by far the simplest was this patch that just increases the size of the CharChunk internal buffer when creating the mark sufficiently that flushBuffer() is never called. With this patch I can't break mark/reset with bug 44494's test case in single or multibyte mode. Oops, I had forgotten that part, there was indeed another problem with that. Sorry. No problem. I learned quite a bit digging around in that code ;) If you can see a better way of fixing this then feel free to propose an alternative patch. Mark - To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
