You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@subversion.apache.org by ry...@netidea.com on 2003/04/01 00:20:28 UTC

SOLVED: Re: Problem with the svn python bindings

Hi,

Change # 5502 changed the interface to svn_fs_file_length() without changing 
fs.py in the python bindings.

The latest tip of subversion crashes when I run cvs2svn.py.  Not sure why, 
older versions dont segfault.  Also, viewcvs segfaults with the latest tip.  
I think I'll be syncing backwards to 5501 to avoid these problems.

Regards,
-ryan


ryan <ry...@netidea.com> said: 

> Hi,
> 
> I've taken most of the weekend to install Subversion.  Several problems and
> solutions:
> 1. if you want to install into /usr/local/stow instead of plain /usr/local,
> you cannot use make install to install the subprojects such as apr,
> apr-util, neon, db4.  Do those seperately instead.  Or use the debian
> packages for everything but neon.
> 2. While make install, do this first:
> Export LD_LIBRARY_PATH=/usr/local/stow/subversion/lib
> Export LIBRARY_PATH=/usr/local/stow/subversion/lib
> 
> This worked for me.
> 
> 3. make swig-py
> - had a hard time finding this out.  I didn¹t read INSTALL in the
> bindings/swig directory :-)
> - make sure you already stow'ed subversion before you make install-swig-py
> 
> 
> Finally, I downloaded a cvs version of viewcvs, but I get this thrown python
> error:
> An Exception Has Occurred
> 
> Python Traceback 
> 
> Traceback (most recent call last):
>   File "/usr/local/viewcvs-1.0-dev/lib/viewcvs.py", line 3037, in main
>     run_viewcvs(server)
>   File "/usr/local/viewcvs-1.0-dev/lib/viewcvs.py", line 3007, in
> run_viewcvs
>     view_diff(request)
>   File "/usr/local/viewcvs-1.0-dev/lib/viewcvs.py", line 2705, in view_diff
>     fp = diffobj.get_pipe()
>   File "/usr/local/stow/subversion/lib/svn-python/svn/fs.py", line 103, in
> get_pipe
>     self.get_files()
>   File "/usr/local/stow/subversion/lib/svn-python/svn/fs.py", line 84, in
> get_files
>     len = file_length(self.root1, self.path1, self.pool)
> TypeError: svn_fs_file_length() takes exactly 4 arguments (3 given)
> 
> I looked in the code, and indeed, this is the case... File_length() is
> giving 3 args, when 4 is expected.
> 
> I looked into it, and it looks like the 3rd argument should be a char * of
> some sort, not the pool which it is now.  I tried a few things, but not
> being a Python expert nor really knowing what I'm doing, I had to give up
> :-)
> 
> I'm not really familiar with SWIG or Python binding, so I'm stuck.
> Hopefully someone out in svn land knows what's going on here.
> 
> PS: svn rules.  Good job.
> 
> Regards,
> -ryan
> 
> 
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
> For additional commands, e-mail: dev-help@subversion.tigris.org
> 
> 



---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: SOLVED: Re: Problem with the svn python bindings

Posted by kf...@collab.net.
ryan <ry...@netidea.com> writes:
> I develeloped my Patch independently of Matt's.  I'm not sure which one is
> more correct since I'm not an expert in SWIG.  I really cant say for
> certain.  But it does look like his does more error checking.  So its
> probably better to go with his.

Thanks.

Looks like Mike Pilato did just that, in revision 5514.

-K

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: SOLVED: Re: Problem with the svn python bindings

Posted by ryan <ry...@netidea.com>.
Hi,

I develeloped my Patch independently of Matt's.  I'm not sure which one is
more correct since I'm not an expert in SWIG.  I really cant say for
certain.  But it does look like his does more error checking.  So its
probably better to go with his.

As for where I 'got stuck', I got stuck with the interface to
svn_stream_read().  Fs.py was passing in 2 parameters, and the function
takes 3 - but 1 of the parameters is a char * buffer out.  I couldn't
resolve that mismatch, so I had to give up.

Regards,
-ryan




On 4/1/03 9:01 AM, "kfogel@collab.net" <kf...@collab.net> wrote:

> ryan <ry...@netidea.com> writes:
>> I spent a bunch of more time looking into it.  Matt is right, the .i
>> files are out of sync with the library interfaces.  From my work on it,
>> I discovered i dont know enough about python or swig to really fix the
>> out of sync Python bindings.  the problem is that the svn_* interface is
>> changing faster than the Python bindings.  My solution was to use
>> version 0.20.1.  but before I gave up I made the following fixes which
>> eased the problem a bit:
>> 
>> Index: svn_fs.i
>> ===================================================================
>> --- svn_fs.i    (revision 5501)
>> +++ svn_fs.i    (working copy)
>> @@ -43,6 +43,10 @@
>>      svn_stream_t **
>>  };
>>  
>> +%apply long long * OUTPUT {
>> +    svn_filesize_t *
>> +};
>> +
>>  /* and this is always an OUT param */
>>  %apply const char **OUTPUT { const char ** };
> 
> Hmmm.  Your patch is different than Matt's patch, but you clearly know
> about Matt's patch since your mail quotes it in response :-).  Can you
> say why to use one rather than the other?  Did Matt's patch not solve
> things for you, or is yours as effective but preferable because
> simpler?
> 
>> But then there is a problem with the interface to svn_stream_read which
>> I couldn't quite resolve, but here is a first stab which converts it to
>> using the proper type:
>> 
>> Index: util.i
>> @@ -94,7 +95,7 @@
>>  /*
>> -----------------------------------------------------------------------
>>     fix up the svn_stream_read() ptr/len arguments
>>  */
>> -%typemap(python, in) (char *buffer, apr_size_t *len) ($*2_type temp) {
>> +%typemap(python, in) (char *buffer, svn_filesize_t *len) ($*2_type
>> temp) {
>>      if (!PyInt_Check($input)) {
>>          PyErr_SetString(PyExc_TypeError,
>>                          "expecting an integer for the buffer size");
>> 
>> 
>> 
>> These are all in subversion/bindings/swig
>> 
>> In any case, I got stuck here, and I gave up and used subversion 0.20.1,
>> and I got ViewCVS working in no time flat.
> 
> Thanks for the patches.  When you say "got stuck", what were the exact
> symptoms you were seeing?
> 
> -K
> 


---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: SOLVED: Re: Problem with the svn python bindings

Posted by kf...@collab.net.
ryan <ry...@netidea.com> writes:
> I spent a bunch of more time looking into it.  Matt is right, the .i
> files are out of sync with the library interfaces.  From my work on it,
> I discovered i dont know enough about python or swig to really fix the
> out of sync Python bindings.  the problem is that the svn_* interface is
> changing faster than the Python bindings.  My solution was to use
> version 0.20.1.  but before I gave up I made the following fixes which
> eased the problem a bit:
> 
> Index: svn_fs.i
> ===================================================================
> --- svn_fs.i    (revision 5501)
> +++ svn_fs.i    (working copy)
> @@ -43,6 +43,10 @@
>      svn_stream_t **
>  };
>  
> +%apply long long * OUTPUT {
> +    svn_filesize_t *
> +};
> +
>  /* and this is always an OUT param */
>  %apply const char **OUTPUT { const char ** };

Hmmm.  Your patch is different than Matt's patch, but you clearly know
about Matt's patch since your mail quotes it in response :-).  Can you
say why to use one rather than the other?  Did Matt's patch not solve
things for you, or is yours as effective but preferable because
simpler?

> But then there is a problem with the interface to svn_stream_read which
> I couldn't quite resolve, but here is a first stab which converts it to
> using the proper type:
> 
> Index: util.i
> @@ -94,7 +95,7 @@
>  /*
> -----------------------------------------------------------------------
>     fix up the svn_stream_read() ptr/len arguments
>  */
> -%typemap(python, in) (char *buffer, apr_size_t *len) ($*2_type temp) {
> +%typemap(python, in) (char *buffer, svn_filesize_t *len) ($*2_type
> temp) {
>      if (!PyInt_Check($input)) {
>          PyErr_SetString(PyExc_TypeError,
>                          "expecting an integer for the buffer size");
> 
> 
> 
> These are all in subversion/bindings/swig
> 
> In any case, I got stuck here, and I gave up and used subversion 0.20.1,
> and I got ViewCVS working in no time flat.

Thanks for the patches.  When you say "got stuck", what were the exact
symptoms you were seeing?

-K

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: SOLVED: Re: Problem with the svn python bindings

Posted by ryan <ry...@netidea.com>.
Hi,

I spent a bunch of more time looking into it.  Matt is right, the .i
files are out of sync with the library interfaces.  From my work on it,
I discovered i dont know enough about python or swig to really fix the
out of sync Python bindings.  the problem is that the svn_* interface is
changing faster than the Python bindings.  My solution was to use
version 0.20.1.  but before I gave up I made the following fixes which
eased the problem a bit:

Index: svn_fs.i
===================================================================
--- svn_fs.i    (revision 5501)
+++ svn_fs.i    (working copy)
@@ -43,6 +43,10 @@
     svn_stream_t **
 };
 
+%apply long long * OUTPUT {
+    svn_filesize_t *
+};
+
 /* and this is always an OUT param */
 %apply const char **OUTPUT { const char ** };
 

But then there is a problem with the interface to svn_stream_read which
I couldn't quite resolve, but here is a first stab which converts it to
using the proper type:

Index: util.i
@@ -94,7 +95,7 @@
 /*
-----------------------------------------------------------------------
    fix up the svn_stream_read() ptr/len arguments
 */
-%typemap(python, in) (char *buffer, apr_size_t *len) ($*2_type temp) {
+%typemap(python, in) (char *buffer, svn_filesize_t *len) ($*2_type
temp) {
     if (!PyInt_Check($input)) {
         PyErr_SetString(PyExc_TypeError,
                         "expecting an integer for the buffer size");



These are all in subversion/bindings/swig

In any case, I got stuck here, and I gave up and used subversion 0.20.1,
and I got ViewCVS working in no time flat.

If anyone wants to know the secrets of installing subversion on debian
with stow, let me know :-)

Regards,
-ryan



Matt Kraai wrote:
> 
> On Mon, Mar 31, 2003 at 10:48:14PM -0500, Garrett Rooney wrote:
> >
> > On Monday, March 31, 2003, at 10:03 PM, kfogel@collab.net wrote:
> >
> > ><ry...@netidea.com> writes:
> > >>Change # 5502 changed the interface to svn_fs_file_length() without
> > >>changing
> > >>fs.py in the python bindings.
> > >
> > >I'm sorry -- I thought those were automatically generated from the
> > >header files, at 'make swig-py' time.  Did I miss a step?
> >
> > it looks like it's the _foo.py stuff that's autogenerated.  the foo.py
> > stuff is hand coded wrappers around the generated code, if i'm reading
> > this right.
> 
> I believe the following patch fixes this.  Would someone please
> test it?
> 
> * subversion/bindings/swig/svn_types.i: Add a typemap for
>   svn_filesize_t.
> 
> Index: subversion/bindings/swig/svn_types.i
> ===================================================================
> --- subversion/bindings/swig/svn_types.i        (revision 5509)
> +++ subversion/bindings/swig/svn_types.i        (working copy)
> @@ -103,6 +103,17 @@
>  */
>  %apply long *OUTPUT { svn_revnum_t * };
> 
> +/* ----------------------------------------------------------------------- */
> +
> +/* Define an OUTPUT typemap for 'svn_filesize_t *'.  For now, we'll
> +   treat it as a 'long' even if that isn't entirely correct...  */
> +
> +%typemap(python,in,numinputs=0) svn_filesize_t * (svn_filesize_t temp)
> +    "$1 = &temp;";
> +
> +%typemap(python,argout,fragment="t_output_helper") svn_filesize_t *
> +    "$result = t_output_helper($result,PyInt_FromLong((long) (*$1)));";
> +
>  /* -----------------------------------------------------------------------
>     Define a general ptr/len typemap. This takes a single script argument
>     and expands it into a ptr/len pair for the native call.
> 
> Matt
> --
> It's most certainly GNU/Linux, not Linux.  Read more at
> http://www.gnu.org/gnu/why-gnu-linux.html.
> 
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
> For additional commands, e-mail: dev-help@subversion.tigris.org

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: SOLVED: Re: Problem with the svn python bindings

Posted by Matt Kraai <kr...@alumni.cmu.edu>.
On Mon, Mar 31, 2003 at 10:48:14PM -0500, Garrett Rooney wrote:
> 
> On Monday, March 31, 2003, at 10:03 PM, kfogel@collab.net wrote:
> 
> ><ry...@netidea.com> writes:
> >>Change # 5502 changed the interface to svn_fs_file_length() without 
> >>changing
> >>fs.py in the python bindings.
> >
> >I'm sorry -- I thought those were automatically generated from the
> >header files, at 'make swig-py' time.  Did I miss a step?
> 
> it looks like it's the _foo.py stuff that's autogenerated.  the foo.py 
> stuff is hand coded wrappers around the generated code, if i'm reading 
> this right.

I believe the following patch fixes this.  Would someone please
test it?

* subversion/bindings/swig/svn_types.i: Add a typemap for
  svn_filesize_t.

Index: subversion/bindings/swig/svn_types.i
===================================================================
--- subversion/bindings/swig/svn_types.i	(revision 5509)
+++ subversion/bindings/swig/svn_types.i	(working copy)
@@ -103,6 +103,17 @@
 */
 %apply long *OUTPUT { svn_revnum_t * };
 
+/* ----------------------------------------------------------------------- */
+
+/* Define an OUTPUT typemap for 'svn_filesize_t *'.  For now, we'll
+   treat it as a 'long' even if that isn't entirely correct...  */
+
+%typemap(python,in,numinputs=0) svn_filesize_t * (svn_filesize_t temp)
+    "$1 = &temp;";
+
+%typemap(python,argout,fragment="t_output_helper") svn_filesize_t *
+    "$result = t_output_helper($result,PyInt_FromLong((long) (*$1)));";
+
 /* -----------------------------------------------------------------------
    Define a general ptr/len typemap. This takes a single script argument
    and expands it into a ptr/len pair for the native call.

Matt
-- 
It's most certainly GNU/Linux, not Linux.  Read more at
http://www.gnu.org/gnu/why-gnu-linux.html.

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: SOLVED: Re: Problem with the svn python bindings

Posted by Garrett Rooney <ro...@electricjellyfish.net>.
On Monday, March 31, 2003, at 10:03 PM, kfogel@collab.net wrote:

> <ry...@netidea.com> writes:
>> Change # 5502 changed the interface to svn_fs_file_length() without 
>> changing
>> fs.py in the python bindings.
>
> I'm sorry -- I thought those were automatically generated from the
> header files, at 'make swig-py' time.  Did I miss a step?

it looks like it's the _foo.py stuff that's autogenerated.  the foo.py 
stuff is hand coded wrappers around the generated code, if i'm reading 
this right.

-garrett


---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: SOLVED: Re: Problem with the svn python bindings

Posted by kf...@collab.net.
<ry...@netidea.com> writes:
> Change # 5502 changed the interface to svn_fs_file_length() without changing 
> fs.py in the python bindings.

I'm sorry -- I thought those were automatically generated from the
header files, at 'make swig-py' time.  Did I miss a step?

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org