[Patch, Java]: Add --enable-reduced-reflection configure option.

[Patch, Java]: Add --enable-reduced-reflection configure option.

[David Daney]

[-- Attachment #1: Type: text/plain, Size: 1464 bytes --]

This patch adds a new configure option to easily allow libgcj to be 
built with the -freduced-reflection option.  This reduces the size (as 
reported by the size command) of libgcj by about 5% on i686-pc-linux. 
Although the resulting library has reduced functionality (as 
serialization, RMI and CORBA don't work with -freduced-reflection), the 
size savings can be important in some circumstances (embedded devices).


The name of the option is --enable-reduced-reflection, although I 
considered and rejected --with-reduced-reflection.


Tested on i686-pc-linux.  Where as expected most -findirect-dispatch 
execution tests failed as well as the interpreter tests and several 
other random tests that rely on accurate reflection meta-data.


OK to commit?

gcc/
2008-08-25  David Daney  <ddaney@avtrex.com>

	* doc/install.texi (--enable-reduced-reflection): Document new option.

libjava/
2008-08-25  David Daney  <ddaney@avtrex.com>

	* configure.ac (reduced-reflection): New AC_ARG_ENABLE.
	(build_libgcj_reduced_reflection): New variable.
	(BUILD_LIBGCJ_REDUCED_REFLECTION): New AM_CONDITIONAL.
	* Makefile.am (LIBGCJ_REDUCED_REFLECTION_FLAGS): New variable.
	(%.lo: %.list): Add LIBGCJ_REDUCED_REFLECTION_FLAGS to compile
	command.
	(java/util/concurrent.lo, java/util/concurrent/atomic.lo,
	java/util/concurrent/locks.lo): Override
	LIBGCJ_REDUCED_REFLECTION_FLAGS.
	* include/Makefile.in, testsuite/Makefile.in, gcj/Makefile.in,
	configure: Regenerate.


[-- Attachment #2: enable-reduced-reflection.patch --]
[-- Type: text/x-patch, Size: 2963 bytes --]

Index: gcc/doc/install.texi
===================================================================
--- gcc/doc/install.texi	(revision 139335)
+++ gcc/doc/install.texi	(working copy)
@@ -1582,6 +1582,13 @@ these options.  This allows the compile-
 dependencies when statically linking to libgcj.  However it makes it
 impossible to override the affected portions of libgcj at run-time.
 
+@item --enable-reduced-reflection
+Build most of libgcj with @option{-freduced-reflection}.  This reduces
+the size of libgcj at the expense of not being able to do accurate
+reflection on the classes it contains.  This option is safe if you
+know that code using libgcj will never use reflection on the standard
+runtime classes in libgcj (including using serialization, RMI or CORBA).
+
 @item --with-ecos
 Enable runtime eCos target support.
 
Index: libjava/configure.ac
===================================================================
--- libjava/configure.ac	(revision 139335)
+++ libjava/configure.ac	(working copy)
@@ -526,6 +526,15 @@ AC_ARG_ENABLE(libgcj-bc,
    fi])
 AM_CONDITIONAL(SUPPRESS_LIBGCJ_BC, test "$suppress_libgcj_bc" = "yes")
 
+build_libgcj_reduced_reflection=no
+AC_ARG_ENABLE(reduced-reflection,
+  AS_HELP_STRING([--enable-reduced-reflection],
+                 [enable or disable(default) -freduced-reflection when building portions of libgcj]),
+  [if test "$enable_reduced_reflection" = "yes"; then
+     build_libgcj_reduced_reflection=yes
+   fi])
+AM_CONDITIONAL(BUILD_LIBGCJ_REDUCED_REFLECTION, test "$build_libgcj_reduced_reflection" = "yes")
+
 # What is the native OS API for MinGW?
 AC_ARG_WITH(win32-nlsapi,
   AS_HELP_STRING([--with-win32-nlsapi=ansi or unicows or unicode],
Index: libjava/Makefile.am
===================================================================
--- libjava/Makefile.am	(revision 139335)
+++ libjava/Makefile.am	(working copy)
@@ -173,6 +173,12 @@ else
 LIBGCJ_BC_FLAGS = -findirect-dispatch -fno-indirect-classes
 endif
 
+if BUILD_LIBGCJ_REDUCED_REFLECTION
+LIBGCJ_REDUCED_REFLECTION_FLAGS = -freduced-reflection
+else
+LIBGCJ_REDUCED_REFLECTION_FLAGS =
+endif
+
 ## Extra CFLAGS used for JNI C sources shared with GNU Classpath.
 PEDANTIC_CFLAGS = -ansi -pedantic -Wall -Wno-long-long
 
@@ -401,8 +407,13 @@ lib-gnu-awt-xlib.la: $(lib_gnu_awt_xlib_
 
 ## Compiling a list of java sources to a single .o.
 
+# Concurrent things use reflection internally.
+java/util/concurrent.lo \
+java/util/concurrent/atomic.lo \
+java/util/concurrent/locks.lo: LIBGCJ_REDUCED_REFLECTION_FLAGS =
+
 %.lo: %.list
-	$(LTGCJCOMPILE) -c -o $@ -fsource-filename=$(here)/classpath/lib/classes -MT $@ -MD -MP -MF $(basename $@).deps @$<
+	$(LTGCJCOMPILE) $(LIBGCJ_REDUCED_REFLECTION_FLAGS) -c -o $@ -fsource-filename=$(here)/classpath/lib/classes -MT $@ -MD -MP -MF $(basename $@).deps @$<
 
 java/lang/Object.lo: classpath/lib/java/lang/Object.class
 	$(LTGCJCOMPILE) -c -o $@ -fsource-filename=$(srcdir)/$(basename $@).java $<

Re: [Patch, Java]: Add --enable-reduced-reflection configure option.

[Andrew Haley]

David Daney wrote:
> This patch adds a new configure option to easily allow libgcj to be
> built with the -freduced-reflection option.  This reduces the size (as
> reported by the size command) of libgcj by about 5% on i686-pc-linux.
> Although the resulting library has reduced functionality (as
> serialization, RMI and CORBA don't work with -freduced-reflection), the
> size savings can be important in some circumstances (embedded devices).
> 
> 
> The name of the option is --enable-reduced-reflection, although I
> considered and rejected --with-reduced-reflection.
> 
> 
> Tested on i686-pc-linux.  Where as expected most -findirect-dispatch
> execution tests failed as well as the interpreter tests and several
> other random tests that rely on accurate reflection meta-data.
> 
> 
> OK to commit?

OK.  I'm a little nervous about this, but I can see the advantages.
I wonder if it might be worth dropping the nonfunctional packages
from libgcj built with reduce reflection.

Andrew.

Re: [Patch, Java]: Add --enable-reduced-reflection configure option.

[David Daney]

Andrew Haley wrote:
> David Daney wrote:
>> This patch adds a new configure option to easily allow libgcj to be
>> built with the -freduced-reflection option.  This reduces the size (as
>> reported by the size command) of libgcj by about 5% on i686-pc-linux.
>> Although the resulting library has reduced functionality (as
>> serialization, RMI and CORBA don't work with -freduced-reflection), the
>> size savings can be important in some circumstances (embedded devices).
>>
>>
>> The name of the option is --enable-reduced-reflection, although I
>> considered and rejected --with-reduced-reflection.
>>
>>
>> Tested on i686-pc-linux.  Where as expected most -findirect-dispatch
>> execution tests failed as well as the interpreter tests and several
>> other random tests that rely on accurate reflection meta-data.
>>
>>
>> OK to commit?
> 
> OK.  I'm a little nervous about this, but I can see the advantages.
> I wonder if it might be worth dropping the nonfunctional packages
> from libgcj built with reduce reflection.
> 

I have thought about this.  And while it may be relatively easy to 
remove RMI and CORBA, there would still be broken serialization which 
would be quite difficult to remove.

I will defer a couple more days while we think about this.

Do we really want this in libgcj?  I don't know, I could keep it as a 
local patch without much pain.

David Daney

Re: [Patch, Java]: Add --enable-reduced-reflection configure option.

[Andrew Haley]

David Daney wrote:
> Andrew Haley wrote:
>> David Daney wrote:
>>> This patch adds a new configure option to easily allow libgcj to be
>>> built with the -freduced-reflection option.  This reduces the size (as
>>> reported by the size command) of libgcj by about 5% on i686-pc-linux.
>>> Although the resulting library has reduced functionality (as
>>> serialization, RMI and CORBA don't work with -freduced-reflection), the
>>> size savings can be important in some circumstances (embedded devices).
>>>
>>>
>>> The name of the option is --enable-reduced-reflection, although I
>>> considered and rejected --with-reduced-reflection.
>>>
>>>
>>> Tested on i686-pc-linux.  Where as expected most -findirect-dispatch
>>> execution tests failed as well as the interpreter tests and several
>>> other random tests that rely on accurate reflection meta-data.
>>>
>>>
>>> OK to commit?
>>
>> OK.  I'm a little nervous about this, but I can see the advantages.
>> I wonder if it might be worth dropping the nonfunctional packages
>> from libgcj built with reduce reflection.
>>
> 
> I have thought about this.  And while it may be relatively easy to
> remove RMI and CORBA, there would still be broken serialization which
> would be quite difficult to remove.
> 
> I will defer a couple more days while we think about this.
> 
> Do we really want this in libgcj?  I don't know, I could keep it as a
> local patch without much pain.

Your patch is OK, and potentially useful to others.  It would be more
useful if we could drop the parts of libgcj that don't work with it,
but that's more like a request for enhancement.

Andrew.

Re: [Patch, Java]: Add --enable-reduced-reflection configure option.

[David Daney]

Andrew Haley wrote:
> David Daney wrote:
>> Andrew Haley wrote:
>>> David Daney wrote:
>>>> This patch adds a new configure option to easily allow libgcj to be
>>>> built with the -freduced-reflection option.  This reduces the size (as
>>>> reported by the size command) of libgcj by about 5% on i686-pc-linux.
>>>> Although the resulting library has reduced functionality (as
>>>> serialization, RMI and CORBA don't work with -freduced-reflection), the
>>>> size savings can be important in some circumstances (embedded devices).
>>>>
>>>>
>>>> The name of the option is --enable-reduced-reflection, although I
>>>> considered and rejected --with-reduced-reflection.
>>>>
>>>>
>>>> Tested on i686-pc-linux.  Where as expected most -findirect-dispatch
>>>> execution tests failed as well as the interpreter tests and several
>>>> other random tests that rely on accurate reflection meta-data.
>>>>
>>>>
>>>> OK to commit?
>>> OK.  I'm a little nervous about this, but I can see the advantages.
>>> I wonder if it might be worth dropping the nonfunctional packages
>>> from libgcj built with reduce reflection.
>>>
>> I have thought about this.  And while it may be relatively easy to
>> remove RMI and CORBA, there would still be broken serialization which
>> would be quite difficult to remove.
>>
>> I will defer a couple more days while we think about this.
>>
>> Do we really want this in libgcj?  I don't know, I could keep it as a
>> local patch without much pain.
> 
> Your patch is OK, and potentially useful to others.  It would be more
> useful if we could drop the parts of libgcj that don't work with it,
> but that's more like a request for enhancement.
> 

Thanks Andrew.

Committed with this corrected ChangeLog:

2008-09-05  David Daney  <ddaney@avtrex.com>

	* doc/install.texi (--enable-reduced-reflection): Document new option.


2008-09-05  David Daney  <ddaney@avtrex.com>

	* configure.ac (reduced-reflection): New AC_ARG_ENABLE.
	(build_libgcj_reduced_reflection): New variable.
	(BUILD_LIBGCJ_REDUCED_REFLECTION): New AM_CONDITIONAL.
	* Makefile.am (LIBGCJ_REDUCED_REFLECTION_FLAGS): New variable.
	(%.lo: %.list): Add LIBGCJ_REDUCED_REFLECTION_FLAGS to compile
	command.
	(java/util/concurrent.lo, java/util/concurrent/atomic.lo,
	java/util/concurrent/locks.lo): Override
	LIBGCJ_REDUCED_REFLECTION_FLAGS.
	* Makefile.in, include/Makefile.in, testsuite/Makefile.in,
	gcj/Makefile.in, configure: Regenerate.