You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@subversion.apache.org by Madan U Sreenivasan <ma...@collab.net> on 2006/10/24 23:36:20 UTC

[PATCH][MERGE-TRACKING] Step 3 of recording copyfrom mergeinfo during repos to repos copy

Hi,

    Please find attached step 3 in recording copyfrom mergeinfo during a  
repos to repos copy. whats pending?

    - Python tests for the same

Regards,
Madan.

Re: [PATCH][MERGE-TRACKING] Step 3 of recording copyfrom mergeinfo during repos to repos copy

Posted by Daniel Rall <dl...@collab.net>.
On Fri, 27 Oct 2006, Kamesh Jayachandran wrote:

> Madan U Sreenivasan wrote:
> >On Thu, 26 Oct 2006 12:31:38 +0530, Kamesh Jayachandran 
> ><ka...@collab.net> wrote:
> >
> >>
> >>>>>+static svn_error_t *
> >>>>>+get_src_merge_info(svn_ra_session_t *ra_session,
> >>>>>+                   apr_hash_t **src_mergeinfo,
> >>>>>+                   const char *src_path,
> >>>>>+                   svn_revnum_t src_rev,
> >>>>>+                   apr_pool_t *pool)
> >>>>>
> >>>>May be we should export this function so that 
> >>>>subversion/libsvn_client/diff.c:get_wc_target_merge_info can make 
> >>>>use of the same. I could see the scope for its usage when we might 
> >>>>have 'mergeaudit' subcommands.
> >>>
> >>>Could you explain a bit more in detail. All this function does is 
> >>>call svn_ra_get_merge_info(), and wraps it up in a mergeinfo hash.
> >>Yes we have a similar code lying out in the function 
> >>'get_wc_target_merge_info' inside subversion/libsvn_client/diff.c. If 
> >>you make get_src_merge_info public(that is what I meant when I said 
> >>export(static functions are not exported) it ) we can reuse it there 
> >>and possible future consumers like subcommands like 'svn mergeaudit'.
> >>May be we can call it svn_client_get_mergeinfo_for_single_path() ??
> >
> >No, these are not the same...
> 
> <snip from libsvn_client/diff.c>
>  APR_ARRAY_PUSH(mergeinfo_paths, const char *) = repos_rel_path;
> 
>  /* ### TODO: To handle sub-tree merge info, the list will need to
>     ### include the those child paths which have merge info which
>     ### differs from that of TARGET_WCPATH, and if those paths are
>     ### directories, their children as well. */
> 
>  SVN_ERR(svn_ra_get_merge_info(ra_session, &repos_mergeinfo, 
> mergeinfo_paths,
>                                target_rev, TRUE, pool));
> //This line is independent of the earlier one and the following one so 
> can be moved above or below.
>  SVN_ERR(parse_merge_info(target_mergeinfo, *entry, target_wcpath,
>                           adm_access, ctx, pool));
> 
> if (repos_mergeinfo != NULL)
>    {
>      apr_hash_t *target_repos_mergeinfo =
>        apr_hash_get(repos_mergeinfo, repos_rel_path, APR_HASH_KEY_STRING);
> 
> 
> </snip>
> <snip from your patch>
> 
>  apr_hash_t *merge_info;
>  apr_array_header_t *rel_paths = apr_array_make(pool, 1,
>                                                 sizeof(src_path));
> 
>  APR_ARRAY_PUSH(rel_paths, const char *) = src_path;
> 
>  SVN_ERR(svn_ra_get_merge_info(ra_session, &merge_info, rel_paths,
>                                src_rev, TRUE, pool));
> 
>  /* Dereference to obtain only the merge info of the src_path provided */
>  if (merge_info)
>    *src_mergeinfo = apr_hash_get(merge_info, src_path, APR_HASH_KEY_STRING);
>  else
>    *src_mergeinfo = NULL;
> 
>  return SVN_NO_ERROR;
> 
> </snip>
> I meant to reuse this implementation in 
> libsvn_client/diff.c:get_wc_target_merge_info.
> 
> Hope I made it clear.

Good suggestion, Kamesh.  I've added an
svn_client__get_merge_info_for_path() API in r22130, which is now used
from get_wc_target_merge_info().  As WC -> repos 'copy' is implemented
in terms of Merge Tracking, we'll likely refactor more of
get_wc_target_merge_info() into the shared subroutine.

I will also use this new API when committing Madan's patch.

Re: [PATCH][MERGE-TRACKING] Step 3 of recording copyfrom mergeinfo during repos to repos copy

Posted by Kamesh Jayachandran <ka...@collab.net>.
Madan U Sreenivasan wrote:
> On Thu, 26 Oct 2006 12:31:38 +0530, Kamesh Jayachandran 
> <ka...@collab.net> wrote:
>
>>
>>>>> +static svn_error_t *
>>>>> +get_src_merge_info(svn_ra_session_t *ra_session,
>>>>> +                   apr_hash_t **src_mergeinfo,
>>>>> +                   const char *src_path,
>>>>> +                   svn_revnum_t src_rev,
>>>>> +                   apr_pool_t *pool)
>>>>>
>>>> May be we should export this function so that 
>>>> subversion/libsvn_client/diff.c:get_wc_target_merge_info can make 
>>>> use of the same. I could see the scope for its usage when we might 
>>>> have 'mergeaudit' subcommands.
>>>
>>> Could you explain a bit more in detail. All this function does is 
>>> call svn_ra_get_merge_info(), and wraps it up in a mergeinfo hash.
>> Yes we have a similar code lying out in the function 
>> 'get_wc_target_merge_info' inside subversion/libsvn_client/diff.c. If 
>> you make get_src_merge_info public(that is what I meant when I said 
>> export(static functions are not exported) it ) we can reuse it there 
>> and possible future consumers like subcommands like 'svn mergeaudit'.
>> May be we can call it svn_client_get_mergeinfo_for_single_path() ??
>
> No, these are not the same...

<snip from libsvn_client/diff.c>
  APR_ARRAY_PUSH(mergeinfo_paths, const char *) = repos_rel_path;

  /* ### TODO: To handle sub-tree merge info, the list will need to
     ### include the those child paths which have merge info which
     ### differs from that of TARGET_WCPATH, and if those paths are
     ### directories, their children as well. */

  SVN_ERR(svn_ra_get_merge_info(ra_session, &repos_mergeinfo, 
mergeinfo_paths,
                                target_rev, TRUE, pool));
//This line is independent of the earlier one and the following one so 
can be moved above or below.
  SVN_ERR(parse_merge_info(target_mergeinfo, *entry, target_wcpath,
                           adm_access, ctx, pool));

 if (repos_mergeinfo != NULL)
    {
      apr_hash_t *target_repos_mergeinfo =
        apr_hash_get(repos_mergeinfo, repos_rel_path, APR_HASH_KEY_STRING);


</snip>
<snip from your patch>

  apr_hash_t *merge_info;
  apr_array_header_t *rel_paths = apr_array_make(pool, 1,
                                                 sizeof(src_path));

  APR_ARRAY_PUSH(rel_paths, const char *) = src_path;

  SVN_ERR(svn_ra_get_merge_info(ra_session, &merge_info, rel_paths,
                                src_rev, TRUE, pool));

  /* Dereference to obtain only the merge info of the src_path provided */
  if (merge_info)
    *src_mergeinfo = apr_hash_get(merge_info, src_path, APR_HASH_KEY_STRING);
  else
    *src_mergeinfo = NULL;

  return SVN_NO_ERROR;

</snip>
I meant to reuse this implementation in 
libsvn_client/diff.c:get_wc_target_merge_info.

Hope I made it clear.
>
>>>>> +{
>>>>> +  /* Dereference to obtain only the merge info of the src_path 
>>>>> provided */
>>>>> +  if (merge_info)
>>>>>
>>>> No need to check for nullity of merge_info. Unless rel_paths->nelts 
>>>> == 0 merge_info will never be null.
>>>
>>> We need to ensure that src_mergeinfo is set to NULL, thats why the 
>>> null-check.
>> No need to ensure that. Across our code, atleast the places where I 
>> have seen, the callee always sets the apr_hash_t*. Checking for NULL 
>> there is redundancy.
>>
>> In this particular case refer 'svn_repos_fs_get_merge_info' in 
>> subversion/libsvn_repos/fs-wrap.c you will realize that 
>> 'src_mergeinfo' will never be NULL.
>
> Please read the code. We also have to extract the inner-hash of the 
> mergeinfo. To do this, we have to know whether the mergeinfo hash is 
> non-NULL before we can call the apr_hash_get() function.
Please read the implementation of the following carefully.
a) 'svn_repos_fs_get_merge_info' in subversion/libsvn_repos/fs-wrap.c
b)svn_fs_merge_info__get_merge_info in 
subversion/libsvn_fs_util/merge-info-sqlite-index.c
You will understand that for this case(you have some meaningful stuff 
inside your rel_paths array) that you work with 'mergeinfo' hash will 
never be null, as with many of our library functions having 
'apr_hash_t**' kind of out params.

With regards
Kamesh Jayachandran

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

Re: [PATCH][MERGE-TRACKING] Step 3 of recording copyfrom mergeinfo during repos to repos copy

Posted by Madan U Sreenivasan <ma...@collab.net>.
On Thu, 26 Oct 2006 12:31:38 +0530, Kamesh Jayachandran  
<ka...@collab.net> wrote:

>
>>>> +static svn_error_t *
>>>> +get_src_merge_info(svn_ra_session_t *ra_session,
>>>> +                   apr_hash_t **src_mergeinfo,
>>>> +                   const char *src_path,
>>>> +                   svn_revnum_t src_rev,
>>>> +                   apr_pool_t *pool)
>>>>
>>> May be we should export this function so that  
>>> subversion/libsvn_client/diff.c:get_wc_target_merge_info can make use  
>>> of the same. I could see the scope for its usage when we might have  
>>> 'mergeaudit' subcommands.
>>
>> Could you explain a bit more in detail. All this function does is call  
>> svn_ra_get_merge_info(), and wraps it up in a mergeinfo hash.
> Yes we have a similar code lying out in the function  
> 'get_wc_target_merge_info' inside subversion/libsvn_client/diff.c. If  
> you make get_src_merge_info public(that is what I meant when I said  
> export(static functions are not exported) it ) we can reuse it there and  
> possible future consumers like subcommands like 'svn mergeaudit'.
> May be we can call it svn_client_get_mergeinfo_for_single_path() ??

No, these are not the same...

>>>> +{
>>>> +  /* Dereference to obtain only the merge info of the src_path  
>>>> provided */
>>>> +  if (merge_info)
>>>>
>>> No need to check for nullity of merge_info. Unless rel_paths->nelts ==  
>>> 0 merge_info will never be null.
>>
>> We need to ensure that src_mergeinfo is set to NULL, thats why the  
>> null-check.
> No need to ensure that. Across our code, atleast the places where I have  
> seen, the callee always sets the apr_hash_t*. Checking for NULL there is  
> redundancy.
>
> In this particular case refer 'svn_repos_fs_get_merge_info' in  
> subversion/libsvn_repos/fs-wrap.c you will realize that 'src_mergeinfo'  
> will never be NULL.

Please read the code. We also have to extract the inner-hash of the  
mergeinfo. To do this, we have to know whether the mergeinfo hash is  
non-NULL before we can call the apr_hash_get() function.

Regards,
Madan.

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

Re: [PATCH][MERGE-TRACKING] Step 3 of recording copyfrom mergeinfo during repos to repos copy

Posted by Kamesh Jayachandran <ka...@collab.net>.
>>> +static svn_error_t *
>>> +get_src_merge_info(svn_ra_session_t *ra_session,
>>> +                   apr_hash_t **src_mergeinfo,
>>> +                   const char *src_path,
>>> +                   svn_revnum_t src_rev,
>>> +                   apr_pool_t *pool)
>>>
>> May be we should export this function so that 
>> subversion/libsvn_client/diff.c:get_wc_target_merge_info can make use 
>> of the same. I could see the scope for its usage when we might have 
>> 'mergeaudit' subcommands.
>
> Could you explain a bit more in detail. All this function does is call 
> svn_ra_get_merge_info(), and wraps it up in a mergeinfo hash.
Yes we have a similar code lying out in the function 
'get_wc_target_merge_info' inside subversion/libsvn_client/diff.c. If 
you make get_src_merge_info public(that is what I meant when I said 
export(static functions are not exported) it ) we can reuse it there and 
possible future consumers like subcommands like 'svn mergeaudit'.
May be we can call it svn_client_get_mergeinfo_for_single_path() ??

>
>>> +{
>>> +  /* Dereference to obtain only the merge info of the src_path 
>>> provided */
>>> +  if (merge_info)
>>>
>> No need to check for nullity of merge_info. Unless rel_paths->nelts 
>> == 0 merge_info will never be null.
>
> We need to ensure that src_mergeinfo is set to NULL, thats why the 
> null-check.
No need to ensure that. Across our code, atleast the places where I have 
seen, the callee always sets the apr_hash_t*. Checking for NULL there is 
redundancy.

In this particular case refer 'svn_repos_fs_get_merge_info' in 
subversion/libsvn_repos/fs-wrap.c you will realize that 'src_mergeinfo' 
will never be NULL.
Anyway I could see this kind of too cautious(????) null checking atleast 
in 2 instances(for which my patch found its place in issue tracker!!!, 
and hence did not give much motivation to dig further.)
>
>>> +  
>>>  +  /* Obtain copyfrom information of the source */
>>>    SVN_ERR(get_copyfrom_merge_info(ra_session, &copyfrom_mergeinfo,
>>>                                    src_rel_path, copyfrom_path,
>>>                                    src_revnum, pool));
>>>
>> What about using only 'get_src_merge_info'?
>
> get_src_merge_info() is to get the existing mergeinfo of the source. 
> get_copyfrom_merge_info() is to get the copyfrom information of the 
> source. How can we use the same function?
>
Oh I misunderstood!.
I thought 'get_implied_merge_info' would get the 'svn:mergeinfo' as 
available on 'copy_from_source_path'. I am wrong it just gives the 
'copy_from_source_path:created_rev-cur_rev'. So we need both this 
mergeinfo and mergeinfo crawl on 'copy_from_mergeinfo'.

With regards
Kamesh Jayachandran

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

Re: [PATCH][MERGE-TRACKING] Step 3 of recording copyfrom mergeinfo during repos to repos copy

Posted by Madan U Sreenivasan <ma...@collab.net>.
On Wed, 25 Oct 2006 18:06:37 +0530, Kamesh Jayachandran  
<ka...@collab.net> wrote:

>
>> * subversion/libsvn_client/copy.c
>>   (includes): svn_mergeinfo.h
>>   (get_src_merge_info): New. Obtains the existing mergeinfo hash for the
>>    given path.
>>   (calculate_target_merge_info): Modfied to use get_src_merge_info(),  
>> and
>>    merge the result with the already available copyfrom information.
>>
> Please use svn diff. I could not see 'calculate_target_merge_info'  
> anywhere in the merge-tracking branch.

calculate_target_merge_info() is part of my uncommitted patch submitted at  
http://svn.haxx.se/dev/archive-2006-10/0681.shtml. Sorry, I should have  
mentioned the dependency.

>> ------------------------------------------------------------------------
>>
>> --- subversion/libsvn_client/copy.c.bu4diff	2006-10-24  
>> 04:21:03.000000000 +0530
>> +++ subversion/libsvn_client/copy.c	2006-10-25 04:57:26.000000000 +0530
>> @@ -31,6 +31,7 @@
>>  #include "svn_opt.h"
>>  #include "svn_time.h"
>>  #include "svn_props.h"
>> +#include "svn_mergeinfo.h"
>>   #include "client.h"
>>  @@ -239,6 +240,33 @@
>>    return SVN_NO_ERROR;
>>  }
>>  +/* Obtain the mergeinfo hash of the given path in SRC_MERGEINFO.
>> +   If theres no merge info available, SRC_MERGEINFO will be NULL */
>> +static svn_error_t *
>> +get_src_merge_info(svn_ra_session_t *ra_session,
>> +                   apr_hash_t **src_mergeinfo,
>> +                   const char *src_path,
>> +                   svn_revnum_t src_rev,
>> +                   apr_pool_t *pool)
>>
> May be we should export this function so that  
> subversion/libsvn_client/diff.c:get_wc_target_merge_info can make use of  
> the same. I could see the scope for its usage when we might have  
> 'mergeaudit' subcommands.

Could you explain a bit more in detail. All this function does is call  
svn_ra_get_merge_info(), and wraps it up in a mergeinfo hash.

>> +{
>> +  apr_hash_t *merge_info;
>> +  apr_array_header_t *rel_paths = apr_array_make(pool, 1,
>> +                                                 sizeof(src_path));
>> +
>> +  APR_ARRAY_PUSH(rel_paths, const char *) = src_path;
>> +
>> +  SVN_ERR(svn_ra_get_merge_info(ra_session, &merge_info, rel_paths,
>> +                                src_rev, TRUE, pool));
>> +
>> +  /* Dereference to obtain only the merge info of the src_path  
>> provided */
>> +  if (merge_info)
>>
> No need to check for nullity of merge_info. Unless rel_paths->nelts == 0  
> merge_info will never be null.

We need to ensure that src_mergeinfo is set to NULL, thats why the  
null-check.

>> +    *src_mergeinfo = apr_hash_get(merge_info, src_path,  
>> APR_HASH_KEY_STRING);
>> +  else
>> +    *src_mergeinfo = NULL;
>> +
>> +  return SVN_NO_ERROR;
>> +}
>> +
>>  /* Obtain the copyfrom merge info and the existing merge info of
>>     the source path, merge them and set as a svn_string_t in
>>     TARGET_MERGEINFO. */
>> @@ -252,7 +280,7 @@
>>  {
>>    const char *repos_root;
>>    const char *copyfrom_path = copyfrom_url;
>> -  apr_hash_t *copyfrom_mergeinfo;
>> +  apr_hash_t *copyfrom_mergeinfo, *src_mergeinfo;
>>    svn_stringbuf_t *mergeinfo;
>>     /* Find src path relative to the repos root */
>> @@ -260,14 +288,19 @@
>>    while (*copyfrom_path++ == *repos_root++);
>>    copyfrom_path--;
>>  +  /* Obtain copyfrom information of the source */
>>    SVN_ERR(get_copyfrom_merge_info(ra_session, &copyfrom_mergeinfo,
>>                                    src_rel_path, copyfrom_path,
>>                                    src_revnum, pool));
>>
> What about using only 'get_src_merge_info'?

get_src_merge_info() is to get the existing mergeinfo of the source.  
get_copyfrom_merge_info() is to get the copyfrom information of the  
source. How can we use the same function?

Thanks for the comments.

Regards,
Madan.

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

Re: [PATCH][MERGE-TRACKING] Step 3 of recording copyfrom mergeinfo during repos to repos copy

Posted by Kamesh Jayachandran <ka...@collab.net>.
> * subversion/libsvn_client/copy.c
>   (includes): svn_mergeinfo.h
>   (get_src_merge_info): New. Obtains the existing mergeinfo hash for the
>    given path.
>   (calculate_target_merge_info): Modfied to use get_src_merge_info(), and
>    merge the result with the already available copyfrom information.
>   
Please use svn diff. I could not see 'calculate_target_merge_info' 
anywhere in the merge-tracking branch.
> ------------------------------------------------------------------------
>
> --- subversion/libsvn_client/copy.c.bu4diff	2006-10-24 04:21:03.000000000 +0530
> +++ subversion/libsvn_client/copy.c	2006-10-25 04:57:26.000000000 +0530
> @@ -31,6 +31,7 @@
>  #include "svn_opt.h"
>  #include "svn_time.h"
>  #include "svn_props.h"
> +#include "svn_mergeinfo.h"
>  
>  #include "client.h"
>  
> @@ -239,6 +240,33 @@
>    return SVN_NO_ERROR;
>  }
>  
> +/* Obtain the mergeinfo hash of the given path in SRC_MERGEINFO.
> +   If theres no merge info available, SRC_MERGEINFO will be NULL */
> +static svn_error_t *
> +get_src_merge_info(svn_ra_session_t *ra_session,
> +                   apr_hash_t **src_mergeinfo,
> +                   const char *src_path,
> +                   svn_revnum_t src_rev,
> +                   apr_pool_t *pool)
>   
May be we should export this function so that 
subversion/libsvn_client/diff.c:get_wc_target_merge_info can make use of 
the same. I could see the scope for its usage when we might have 
'mergeaudit' subcommands.
> +{
> +  apr_hash_t *merge_info;
> +  apr_array_header_t *rel_paths = apr_array_make(pool, 1,
> +                                                 sizeof(src_path));
> +
> +  APR_ARRAY_PUSH(rel_paths, const char *) = src_path;
> +
> +  SVN_ERR(svn_ra_get_merge_info(ra_session, &merge_info, rel_paths,
> +                                src_rev, TRUE, pool));
> +
> +  /* Dereference to obtain only the merge info of the src_path provided */
> +  if (merge_info)
>   
No need to check for nullity of merge_info. Unless rel_paths->nelts == 0 
merge_info will never be null.
> +    *src_mergeinfo = apr_hash_get(merge_info, src_path, APR_HASH_KEY_STRING);
> +  else
> +    *src_mergeinfo = NULL;
> +
> +  return SVN_NO_ERROR;
> +}
> +
>  /* Obtain the copyfrom merge info and the existing merge info of
>     the source path, merge them and set as a svn_string_t in
>     TARGET_MERGEINFO. */
> @@ -252,7 +280,7 @@
>  {
>    const char *repos_root;
>    const char *copyfrom_path = copyfrom_url;
> -  apr_hash_t *copyfrom_mergeinfo;
> +  apr_hash_t *copyfrom_mergeinfo, *src_mergeinfo;
>    svn_stringbuf_t *mergeinfo;
>  
>    /* Find src path relative to the repos root */
> @@ -260,14 +288,19 @@
>    while (*copyfrom_path++ == *repos_root++);
>    copyfrom_path--;
>  
> +  /* Obtain copyfrom information of the source */
>    SVN_ERR(get_copyfrom_merge_info(ra_session, &copyfrom_mergeinfo,
>                                    src_rel_path, copyfrom_path,
>                                    src_revnum, pool));
>   
What about using only 'get_src_merge_info'?
With regards
Kamesh Jayachandran

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

Re: [PATCH][MERGE-TRACKING] Step 3 of recording copyfrom mergeinfo during repos to repos copy

Posted by Daniel Rall <dl...@collab.net>.
On Wed, 25 Oct 2006, Madan S. wrote:
...
> Step 3 of recording copyfrom history as mergeinfo on repos to repos copy.
> This step
>  - obtains the existing merge info on the source path
>  - Merges it with the already obtained copyfrom information
> 
> On the merge-tracking branch:
> 
> * subversion/libsvn_client/copy.c
>   (includes): svn_mergeinfo.h
>   (get_src_merge_info): New. Obtains the existing mergeinfo hash for the
>    given path.
>   (calculate_target_merge_info): Modfied to use get_src_merge_info(), and
>    merge the result with the already available copyfrom information.

I've committed a tweaked version of this patch in r22131.