2017-01-26 15:29 GMT+01:00 Mark Thomas <ma...@apache.org>:
> On 26 January 2017 13:53:22 GMT+00:00, Konstantin Kolinko <
> knst.koli...@gmail.com> wrote:
> >2017-01-26 16:00 GMT+03:00 Rémy Maucherat <r...@apache.org>:
> >> 2017-01-26 12:10 GMT+01:00 Mark Thomas <ma...@apache.org>:
> >>
> >>> - One TCK test fails because an associated JSP no longer compiles
> >>> because it reaches the 64k method limit. The JSP has a very large
> >>> number of nested tags.
> >>>
> >>> I'm going to take another look at the JSP Generation to see if there
> >is
> >>> anything we can do to make even a small improvement.
> >>>
> >>> :( It could be an option to simply revert the try/finally change in
> >all
> >> branches except 8.5 and 9, after all nobody complained about it.
> >
> >1. I think that try/finally change fixed a real bug. It is better to
> >keep it.
> >
> >
> >2. Looking at java code generated for JSPs by 6.0.50, the additional
> >code is in helper methods ("_jspx_meth_"*), not in the main big
> >_jspService() method.
>
> If you use tags with bodies that contain scriptlets then the code appears
> in the main service method.
>
> >There is not much of that additional code. Just a try/catch and a
> >boolean variable.
> >So I also do not see what can be improved.
> >
> >A "catch (Throwable t)" block at the end of _jspService() could be
> >extracted into a helper method as non-trivial code, but it is only a
> >dozen of lines and it is present only once.
>
> We might be able to save a few bytes with some helper methods but it won't
> be much, if any.
>
> >It needs some actual numbers - how big of a JSP can be compiled by
> >Tomcat.
>
> It took around 250 tags with a scriplet on a single page to trigger the
> problem. I need to test how much difference the try/finally fix made.
>
Yes, the problem is that once you add a scriptlet, there can be a variable
declaration in it (well, of course there is, people don't use scriptlets to
do clean stuff) so there needs to be full visibility of it to build. We had
been trying to remove scriptlets for w very long time, it never worked, we
still have people immediately asking for support of the new Java language
features in JSPs, so brand new code being written right now still features
scriptlets.
Also sad, the original size limits of the JVM were rather low, and never
changed.
Sorry for the complaints ...
Rmy
>
> >3. There are some Jasper options that produce a more compact java
> >code. Especially the option to generates a single out.write() call for
> >a sequence of text instead of separate call for each line of text. A
> >bugzilla report mentions it.
> >
> > <init-param>
> > <param-name>mappedfile</param-name>
> > <param-value>false</param-value>
> > </init-param>
> >
> >http://tomcat.apache.org/tomcat-6.0-doc/jasper-howto.html
> >
> >I wonder whether the mentioned TCK test will pass with that option.
>
> I'll take a look.
>
> Mark
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: dev-unsubscr...@tomcat.apache.org
> For additional commands, e-mail: dev-h...@tomcat.apache.org
>
>
