[PATCH] libjava: Add option to disable BC ABI in libgcj.

[PATCH] libjava: Add option to disable BC ABI in libgcj.

[David Daney]

[-- Attachment #1: Type: text/plain, Size: 1274 bytes --]

Static linking with libgcj does not work well when portions are compiled
with the BC ABI.  The problem is that the compile time linker is not
able to resolve all the needed dependencies.

This patch adds a new configure option (--disable-libgcj-bc) that forces
all code in libgcj to be compiled with the C++ ABI.  The default is to
compile libgcj 'normally'.  This option would only be used by those
statically linking to libgcj.

Tested on i686-pc-linux-gnu with no failures in libjava testsuite.  I
also verified by inspecting build output that -findirect-dispatch
-fno-indirect-classes is being passed when compiling the BC ABI files.

OK to commit?

gcc/
2007-12-21  David Daney  <ddaney@avtrex.com>

    * doc/install.texi (disable-libgcj-bc): Document new option.

libjava/
2007-12-21  David Daney  <ddaney@avtrex.com>

    * scripts/makemake.tcl (emit_bc_rule): Use $(LIBGCJ_BC_FLAGS)
    instead of -findirect-dispatch -fno-indirect-classes.
    * configure.ac (libgcj-bc): New AC_ARG_ENABLE.
    (SUPPRESS_LIBGCJ_BC): New AM_CONDITIONAL.
    * Makefile.am (LIBGCJ_BC_FLAGS): New variable.
    * Makefile.in: Regenerate.
    * include/Makefile.in: Same.
    * testsuite/Makefile.in: Same.
    * configure: Same.
    * gcj/Makefile.in: Same.
    * sources.am: Same.


[-- Attachment #2: libgcj-bc.diff --]
[-- Type: text/x-patch, Size: 2681 bytes --]

Index: gcc/doc/install.texi
===================================================================
--- gcc/doc/install.texi	(revision 131121)
+++ gcc/doc/install.texi	(working copy)
@@ -1554,6 +1554,16 @@ using non-functional stubs for native me
 @item --disable-jvmpi
 Disable JVMPI support.
 
+@item --disable-libgcj-bc
+Disable BC ABI compilation of certain parts of libgcj.  By default,
+some portions of libgcj are compiled with @option{-findirect-dispatch}
+@option{-fno-indirect-classes}.  This allows them to be overridden at
+runtime.
+
+If @option{--disable-libgcj-bc} is specified, libgcj is built without
+these options.  This makes it impossible to override portions of
+libgcj at runtime, but can make it easier to statically link to libgcj.
+
 @item --with-ecos
 Enable runtime eCos target support.
 
Index: libjava/scripts/makemake.tcl
===================================================================
--- libjava/scripts/makemake.tcl	(revision 131121)
+++ libjava/scripts/makemake.tcl	(working copy)
@@ -317,7 +317,9 @@ proc emit_bc_rule {package} {
   if {$package_map($package) == "bc"} {
     puts -nonewline "-fjni "
   }
-  puts "-findirect-dispatch -fno-indirect-classes -c -o $loname @$tname"
+  # Unless bc is disabled with --disable-libgcj-bc, $(LIBGCJ_BC_FLAGS) is:
+  #   -findirect-dispatch -fno-indirect-classes
+  puts "\$(LIBGCJ_BC_FLAGS) -c -o $loname @$tname"
   puts "\t@rm -f $tname"
   puts ""
 
Index: libjava/configure.ac
===================================================================
--- libjava/configure.ac	(revision 131121)
+++ libjava/configure.ac	(working copy)
@@ -514,6 +514,15 @@ AC_ARG_WITH(java-home,
 AM_CONDITIONAL(JAVA_HOME_SET, test ! -z "$JAVA_HOME")
 AC_SUBST(JAVA_HOME)
 
+suppress_libgcj_bc=no
+AC_ARG_ENABLE(libgcj-bc,
+  AS_HELP_STRING([--enable-libgcj-bc],
+                 [enable(default) or disable BC ABI for portions of libgcj]),
+  [if test "$enable_libgcj_bc" = "no"; then
+     suppress_libgcj_bc=yes
+   fi])
+AM_CONDITIONAL(SUPPRESS_LIBGCJ_BC, test "$suppress_libgcj_bc" = "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 131121)
+++ libjava/Makefile.am	(working copy)
@@ -155,6 +155,12 @@ if USING_GCC
 AM_CFLAGS += $(WARNINGS)
 endif
 
+if SUPPRESS_LIBGCJ_BC
+LIBGCJ_BC_FLAGS =
+else
+LIBGCJ_BC_FLAGS = -findirect-dispatch -fno-indirect-classes
+endif
+
 ## Extra CFLAGS used for JNI C sources shared with GNU Classpath.
 PEDANTIC_CFLAGS = -ansi -pedantic -Wall -Wno-long-long
 

Re: [PATCH] libjava: Add option to disable BC ABI in libgcj.

[Tom Tromey]

>>>>> "David" == David Daney <ddaney@avtrex.com> writes:

David> This patch adds a new configure option (--disable-libgcj-bc) that forces
David> all code in libgcj to be compiled with the C++ ABI.  The default is to
David> compile libgcj 'normally'.  This option would only be used by those
David> statically linking to libgcj.

Ok.  Thanks.

I don't think I can approve the install.texi change.

Tom

Re: [PATCH] libjava: Add option to disable BC ABI in libgcj.

[Andrew Haley]

David Daney writes:
 > Static linking with libgcj does not work well when portions are compiled
 > with the BC ABI.  The problem is that the compile time linker is not
 > able to resolve all the needed dependencies.
 > 
 > This patch adds a new configure option (--disable-libgcj-bc) that forces
 > all code in libgcj to be compiled with the C++ ABI.  The default is to
 > compile libgcj 'normally'.  This option would only be used by those
 > statically linking to libgcj.
 > 
 > Tested on i686-pc-linux-gnu with no failures in libjava testsuite.  I
 > also verified by inspecting build output that -findirect-dispatch
 > -fno-indirect-classes is being passed when compiling the BC ABI files.
 > 
 > OK to commit?
 > 
 > gcc/
 > 2007-12-21  David Daney  <ddaney@avtrex.com>
 > 
 >     * doc/install.texi (disable-libgcj-bc): Document new option.
 > 
 > libjava/
 > 2007-12-21  David Daney  <ddaney@avtrex.com>
 > 
 >     * scripts/makemake.tcl (emit_bc_rule): Use $(LIBGCJ_BC_FLAGS)
 >     instead of -findirect-dispatch -fno-indirect-classes.
 >     * configure.ac (libgcj-bc): New AC_ARG_ENABLE.
 >     (SUPPRESS_LIBGCJ_BC): New AM_CONDITIONAL.
 >     * Makefile.am (LIBGCJ_BC_FLAGS): New variable.
 >     * Makefile.in: Regenerate.
 >     * include/Makefile.in: Same.
 >     * testsuite/Makefile.in: Same.
 >     * configure: Same.
 >     * gcj/Makefile.in: Same.
 >     * sources.am: Same.

This is fine, thanks.

Andrew.

-- 
Red Hat UK Ltd, Amberley Place, 107-111 Peascod Street, Windsor, Berkshire, SL4 1TE, UK
Registered in England and Wales No. 3798903

Re: [PATCH] libjava: Add option to disable BC ABI in libgcj.

[Gerald Pfeifer]

On Fri, 21 Dec 2007, Tom Tromey wrote:
> I don't think I can approve the install.texi change.

Sure you can. ;-)  (Certainly for or those areas where you are a 
maintainer or otherwise authoritative.)

On Fri, 21 Dec 2007, David Daney wrote:
>     * doc/install.texi (disable-libgcj-bc): Document new option.

+@item --disable-libgcj-bc
+Disable BC ABI compilation of certain parts of libgcj.  By default,
+some portions of libgcj are compiled with @option{-findirect-dispatch}
+@option{-fno-indirect-classes}.  This allows them to be overridden at
+runtime.

When first reading this, I got a bit confused by two aspects.  Perhaps 
this is something to address in a follow-up patch, David?

One was that "BC API compilation" (including myself ;-) isn't something 
most readers of doc/install.texi would be familiar with.  How about
expanding this or adding a reference/cross-reference?

The second one was that I wasn't sure what "This" referred to: the 
configure option, or the previous sentence explaining BC ABI compilation.
Perhaps replacing this by ", which" might be more clear?

+If @option{--disable-libgcj-bc} is specified, libgcj is built without
+these options.  This makes it impossible to override portions of
+libgcj at runtime, but can make it easier to statically link to libgcj.

Why does this read "can make" and not just "makes"?  I'm sure there is
some technical background here which I'm missing. :-)  Perhaps a (cross)
reference to more detailed information somewhere else could clarify that?

Gerald

Re: [PATCH] libjava: Add option to disable BC ABI in libgcj.

[David Daney]

[-- Attachment #1: Type: text/plain, Size: 1690 bytes --]

Gerald Pfeifer wrote:
> +@item --disable-libgcj-bc
> +Disable BC ABI compilation of certain parts of libgcj.  By default,
> +some portions of libgcj are compiled with @option{-findirect-dispatch}
> +@option{-fno-indirect-classes}.  This allows them to be overridden at
> +runtime.
>
> When first reading this, I got a bit confused by two aspects.  Perhaps 
> this is something to address in a follow-up patch, David?
>
> One was that "BC API compilation" (including myself ;-) isn't something 
> most readers of doc/install.texi would be familiar with.  How about
> expanding this or adding a reference/cross-reference?
>
> The second one was that I wasn't sure what "This" referred to: the 
> configure option, or the previous sentence explaining BC ABI compilation.
> Perhaps replacing this by ", which" might be more clear?
>
> +If @option{--disable-libgcj-bc} is specified, libgcj is built without
> +these options.  This makes it impossible to override portions of
> +libgcj at runtime, but can make it easier to statically link to libgcj.
>
> Why does this read "can make" and not just "makes"?  I'm sure there is
> some technical background here which I'm missing. :-)  Perhaps a (cross)
> reference to more detailed information somewhere else could clarify that?
>   
How about this version instead?  I think it gets rid of some of the
ambiguity and explains things a bit better.

There really is nothing other than e-mail archives that could be cross
referenced to provide deeper information about this option.

Tested with make info and make html with no errors.

2008-01-25  David Daney  <ddaney@avtrex.com>

    * doc/install.texi (--disable-libgcj-bc):  Reword documentation.


[-- Attachment #2: libgcj-bc-doc.diff --]
[-- Type: text/x-patch, Size: 1018 bytes --]

Index: doc/install.texi
===================================================================
--- doc/install.texi	(revision 131856)
+++ doc/install.texi	(working copy)
@@ -1557,12 +1557,13 @@ Disable JVMPI support.
 @item --disable-libgcj-bc
 Disable BC ABI compilation of certain parts of libgcj.  By default,
 some portions of libgcj are compiled with @option{-findirect-dispatch}
-@option{-fno-indirect-classes}.  This allows them to be overridden at
-runtime.
+and @option{-fno-indirect-classes}, allowing them to be overridden at
+run-time.
 
 If @option{--disable-libgcj-bc} is specified, libgcj is built without
-these options.  This makes it impossible to override portions of
-libgcj at runtime, but can make it easier to statically link to libgcj.
+these options.  This allows the compile-time linker to resolve
+dependencies when statically linking to libgcj.  However it makes it
+impossible to override the effected portions of libgcj at run-time.
 
 @item --with-ecos
 Enable runtime eCos target support.

Re: [PATCH] libjava: Add option to disable BC ABI in libgcj.

[Andrew Haley]

David Daney wrote:
> Gerald Pfeifer wrote:
>> +@item --disable-libgcj-bc
>> +Disable BC ABI compilation of certain parts of libgcj.  By default,
>> +some portions of libgcj are compiled with @option{-findirect-dispatch}
>> +@option{-fno-indirect-classes}.  This allows them to be overridden at
>> +runtime.
>>
>> When first reading this, I got a bit confused by two aspects.  Perhaps 
>> this is something to address in a follow-up patch, David?
>>
>> One was that "BC API compilation" (including myself ;-) isn't something 
>> most readers of doc/install.texi would be familiar with.  How about
>> expanding this or adding a reference/cross-reference?
>>
>> The second one was that I wasn't sure what "This" referred to: the 
>> configure option, or the previous sentence explaining BC ABI compilation.
>> Perhaps replacing this by ", which" might be more clear?
>>
>> +If @option{--disable-libgcj-bc} is specified, libgcj is built without
>> +these options.  This makes it impossible to override portions of
>> +libgcj at runtime, but can make it easier to statically link to libgcj.
>>
>> Why does this read "can make" and not just "makes"?  I'm sure there is
>> some technical background here which I'm missing. :-)  Perhaps a (cross)
>> reference to more detailed information somewhere else could clarify that?
>>   
> How about this version instead?  I think it gets rid of some of the
> ambiguity and explains things a bit better.
> 
> There really is nothing other than e-mail archives that could be cross
> referenced to provide deeper information about this option.

See ftp://gcc.gnu.org/pub/gcc/summit/2004/GCJ%20New%20ABI.pdf

> Tested with make info and make html with no errors.
> 
> 2008-01-25  David Daney  <ddaney@avtrex.com>
> 
>     * doc/install.texi (--disable-libgcj-bc):  Reword documentation.
> 
> 
> 
> ------------------------------------------------------------------------
> 
> Index: doc/install.texi
> ===================================================================
> --- doc/install.texi	(revision 131856)
> +++ doc/install.texi	(working copy)
> @@ -1557,12 +1557,13 @@ Disable JVMPI support.
>  @item --disable-libgcj-bc
>  Disable BC ABI compilation of certain parts of libgcj.  By default,
>  some portions of libgcj are compiled with @option{-findirect-dispatch}
> -@option{-fno-indirect-classes}.  This allows them to be overridden at
> -runtime.
> +and @option{-fno-indirect-classes}, allowing them to be overridden at
> +run-time.
>  
>  If @option{--disable-libgcj-bc} is specified, libgcj is built without
> -these options.  This makes it impossible to override portions of
> -libgcj at runtime, but can make it easier to statically link to libgcj.
> +these options.  This allows the compile-time linker to resolve
> +dependencies when statically linking to libgcj.  However it makes it
> +impossible to override the effected portions of libgcj at run-time.

Is that *a*ffected rather than *e*ffected?

Andrew.

Re: [PATCH] libjava: Add option to disable BC ABI in libgcj.

[David Daney]

Andrew Haley wrote:
> David Daney wrote:
>> Gerald Pfeifer wrote:
>>> +@item --disable-libgcj-bc
>>> +Disable BC ABI compilation of certain parts of libgcj.  By default,
>>> +some portions of libgcj are compiled with @option{-findirect-dispatch}
>>> +@option{-fno-indirect-classes}.  This allows them to be overridden at
>>> +runtime.
>>>
>>> When first reading this, I got a bit confused by two aspects.  
>>> Perhaps this is something to address in a follow-up patch, David?
>>>
>>> One was that "BC API compilation" (including myself ;-) isn't 
>>> something most readers of doc/install.texi would be familiar with.  
>>> How about
>>> expanding this or adding a reference/cross-reference?
>>>
>>> The second one was that I wasn't sure what "This" referred to: the 
>>> configure option, or the previous sentence explaining BC ABI 
>>> compilation.
>>> Perhaps replacing this by ", which" might be more clear?
>>>
>>> +If @option{--disable-libgcj-bc} is specified, libgcj is built without
>>> +these options.  This makes it impossible to override portions of
>>> +libgcj at runtime, but can make it easier to statically link to 
>>> libgcj.
>>>
>>> Why does this read "can make" and not just "makes"?  I'm sure there is
>>> some technical background here which I'm missing. :-)  Perhaps a 
>>> (cross)
>>> reference to more detailed information somewhere else could clarify 
>>> that?
>>>   
>> How about this version instead?  I think it gets rid of some of the
>> ambiguity and explains things a bit better.
>>
>> There really is nothing other than e-mail archives that could be cross
>> referenced to provide deeper information about this option.
>
> See ftp://gcc.gnu.org/pub/gcc/summit/2004/GCJ%20New%20ABI.pdf

While I could certainly reference this paper, I am not sure that 
install.texi is the best place for such a reference.

I was thinking that if someone wanted to know the gory details, they 
could look up -findirect-dispatch.  I was trying to keep the install 
documentation as small as possible while still describing what the new 
option does.

>
>> Tested with make info and make html with no errors.
>>
>> 2008-01-25  David Daney  <ddaney@avtrex.com>
>>
>>     * doc/install.texi (--disable-libgcj-bc):  Reword documentation.
>>
>>
>>
>> ------------------------------------------------------------------------
>>
>> Index: doc/install.texi
>> ===================================================================
>> --- doc/install.texi    (revision 131856)
>> +++ doc/install.texi    (working copy)
>> @@ -1557,12 +1557,13 @@ Disable JVMPI support.
>>  @item --disable-libgcj-bc
>>  Disable BC ABI compilation of certain parts of libgcj.  By default,
>>  some portions of libgcj are compiled with @option{-findirect-dispatch}
>> -@option{-fno-indirect-classes}.  This allows them to be overridden at
>> -runtime.
>> +and @option{-fno-indirect-classes}, allowing them to be overridden at
>> +run-time.
>>  
>>  If @option{--disable-libgcj-bc} is specified, libgcj is built without
>> -these options.  This makes it impossible to override portions of
>> -libgcj at runtime, but can make it easier to statically link to libgcj.
>> +these options.  This allows the compile-time linker to resolve
>> +dependencies when statically linking to libgcj.  However it makes it
>> +impossible to override the effected portions of libgcj at run-time.
>
> Is that *a*ffected rather than *e*ffected?
>
I don't think so, but I have a limited grasp of the English language.  I 
will defer to others on which of 'affected' or 'effected' should be used 
here.

David Daney

Re: [PATCH] libjava: Add option to disable BC ABI in libgcj.

[Andrew Haley]

David Daney wrote:
> Andrew Haley wrote:
>> David Daney wrote:
>>> Gerald Pfeifer wrote:
>>>> +@item --disable-libgcj-bc
>>> There really is nothing other than e-mail archives that could be cross
>>> referenced to provide deeper information about this option.
>>
>> See ftp://gcc.gnu.org/pub/gcc/summit/2004/GCJ%20New%20ABI.pdf
> 
> While I could certainly reference this paper, I am not sure that 
> install.texi is the best place for such a reference.
> 
> I was thinking that if someone wanted to know the gory details, they 
> could look up -findirect-dispatch.  I was trying to keep the install 
> documentation as small as possible while still describing what the new 
> option does.

I agree; I was just answering the "There really is nothing other" point.

>>>  If @option{--disable-libgcj-bc} is specified, libgcj is built without
>>> -these options.  This makes it impossible to override portions of
>>> -libgcj at runtime, but can make it easier to statically link to libgcj.
>>> +these options.  This allows the compile-time linker to resolve
>>> +dependencies when statically linking to libgcj.  However it makes it
>>> +impossible to override the effected portions of libgcj at run-time.
>>
>> Is that *a*ffected rather than *e*ffected?
>>
> I don't think so, but I have a limited grasp of the English language.  I 
> will defer to others on which of 'affected' or 'effected' should be used 
> here.

OK, so take it from me: it's "affected".

Andrew.

Re: [PATCH] libjava: Add option to disable BC ABI in libgcj.

[Gerald Pfeifer]

On Sat, 26 Jan 2008, David Daney wrote:
>> See ftp://gcc.gnu.org/pub/gcc/summit/2004/GCJ%20New%20ABI.pdf
> While I could certainly reference this paper, I am not sure that install.texi
> is the best place for such a reference.
> 
> I was thinking that if someone wanted to know the gory details, they could
> look up -findirect-dispatch.  I was trying to keep the install documentation
> as small as possible while still describing what the new option does.

Fair enough.  Should that link be added to gcj.texi, then?

On Sun, 27 Jan 2008, Andrew Haley wrote:
>> I don't think so, but I have a limited grasp of the English language.  I
>> will defer to others on which of 'affected' or 'effected' should be used
>> here.
> OK, so take it from me: it's "affected".

;-)  David, I assume Andrew is going to give the final Ack for the patch 
(or has done so already).  Just make sure you won't stall on me, I was 
mainly trying to give some feedback.

Gerald

Re: [PATCH] libjava: Add option to disable BC ABI in libgcj.

[David Daney]

[-- Attachment #1: Type: text/plain, Size: 734 bytes --]

Gerald Pfeifer wrote:
> On Sun, 27 Jan 2008, Andrew Haley wrote:
>   
>>> I don't think so, but I have a limited grasp of the English language.  I
>>> will defer to others on which of 'affected' or 'effected' should be used
>>> here.
>>>       
>> OK, so take it from me: it's "affected".
>>     
>
> ;-)  David, I assume Andrew is going to give the final Ack for the patch 
> (or has done so already).  Just make sure you won't stall on me, I was 
> mainly trying to give some feedback.
>   
OK.

Andrew, it's in your hands now.  How about this version?

The only change is to use 'affected' instead off 'effected'.

2008-01-27  David Daney  <ddaney@avtrex.com>

    * doc/install.texi (--disable-libgcj-bc):  Reword documentation.


[-- Attachment #2: libgcj-bc-doc.diff --]
[-- Type: text/x-patch, Size: 1018 bytes --]

Index: doc/install.texi
===================================================================
--- doc/install.texi	(revision 131882)
+++ doc/install.texi	(working copy)
@@ -1557,12 +1557,13 @@ Disable JVMPI support.
 @item --disable-libgcj-bc
 Disable BC ABI compilation of certain parts of libgcj.  By default,
 some portions of libgcj are compiled with @option{-findirect-dispatch}
-@option{-fno-indirect-classes}.  This allows them to be overridden at
-runtime.
+and @option{-fno-indirect-classes}, allowing them to be overridden at
+run-time.
 
 If @option{--disable-libgcj-bc} is specified, libgcj is built without
-these options.  This makes it impossible to override portions of
-libgcj at runtime, but can make it easier to statically link to libgcj.
+these options.  This allows the compile-time linker to resolve
+dependencies when statically linking to libgcj.  However it makes it
+impossible to override the affected portions of libgcj at run-time.
 
 @item --with-ecos
 Enable runtime eCos target support.

Re: [PATCH] libjava: Add option to disable BC ABI in libgcj.

[Andrew Haley]

David Daney wrote:
> Gerald Pfeifer wrote:
>> On Sun, 27 Jan 2008, Andrew Haley wrote:
>>   
>>>> I don't think so, but I have a limited grasp of the English language.  I
>>>> will defer to others on which of 'affected' or 'effected' should be used
>>>> here.
>>>>       
>>> OK, so take it from me: it's "affected".
>>>     
>> ;-)  David, I assume Andrew is going to give the final Ack for the patch 
>> (or has done so already).  Just make sure you won't stall on me, I was 
>> mainly trying to give some feedback.
>>   
> OK.
> 
> Andrew, it's in your hands now.  How about this version?
> 
> The only change is to use 'affected' instead off 'effected'.
> 

OK!  It's OK, really!  Please don't ask again,  :-)

Cheers,
Andrew.