Mattias Engdegård mentioned [1] that nobody has run the 'po-update.sh' script
in trunk for some time, as it complains about some format strings it can't
handle:
../libsvn_fs_fs/index.c:594: ... string ends in the middle of a directive.
../libsvn_fs_fs/index.c:654: ... string ends in the middle of a directive.
../libsvn_fs_fs/cached_data.c:914: ... string ends in the middle of a directive.
The code in each case is similar to:
svn_error_createf(... _("... %" APR_UINT64_T_FMT "..."),
some_uint64_t_value);
The stupid thing is we have no easy and general solution at the moment.
Ideas:
1. Use PRIu64 etc. from <inttypes.h>.
2. Use a helper function to format an integer as a string; insert the result
with "%s".
3. Double-formatting, using "...%%%s..." to insert the correct format specifier.
4. Modify 'gettext' to expand or recognize these macros.
5. Pre-process the source to expand these macros.
1. GNU gettext supports the C'99 PRI... macros from <inttypes.h>:
#include <inttypes.h>
svn_error_createf(..., _("... %" PRIu64 "..."),
some_uint64_t_value);
Limitations:
* GNU gettext is hard-coded to recognize the specific types that are
defined in <inttypes.h>, so it will not work for other similar
definitions of our own or from APR such as APR_OFF_T_FMT.
* We'd have to arrange for <inttypes.h> or equivalent to be
available on all platforms. (That would be useful anyway, and is not
fundamentally difficult.)
2. Use a helper function. (We do this in some places.)
#define ui64toa(pool, i) apr_psprintf(pool, "%"APR_UINT64_T_FMT, i)
svn_error_createf(..., _("... %s..."),
ui64toa(pool, some_uint64_t_value));
Limitations:
* Requires a pool. Some of the current functions don't have one. We
could add one, but it's seems a bit stupid to have to do this, when the
'printf' code already has access to one.
An alternative:
#define ui64toa(i) apr_psprintf(svn_pool_create(NULL), "%"APR_UINT64_T_FMT, i)
svn_error_createf(..., _("... %s..."),
ui64toa(some_uint64_t_value));
But that doesn't release the allocated memory.
3. Double-formatting. (We do this in some places.)
svn_error_createf(...,
apr_psprintf(pool, _("... %%%s ..."),APR_INT64_T_FMT),
some_uint64_t_value);
Limitations:
* Prevents compile-time checking of argument types, and not just for the
arguments that need special handling but for all the arguments.
* Requires a pool.
* Verbose.
4. Modify gettext.
This could be the best long-term solution. However, I'm not tackling it now.
5. Pre-process the source.
I tried this and it worked.
We can simplify the source code as in the following examples. (The first hunk
fixes one of the present problems.)
[[[
Index: subversion/libsvn_fs_fs/cached_data.c
===================================================================
--- subversion/libsvn_fs_fs/cached_data.c (revision 1589948)
+++ subversion/libsvn_fs_fs/cached_data.c (working copy)
@@ -913,6 +913,6 @@
return svn_error_createf(SVN_ERR_REPOS_CORRUPTED, NULL,
- _("No representation found at offset %s "
- "for item %" APR_UINT64_T_FMT
+ _("No representation found at offset %"
APR_OFF_T_FMT
+ " for item %" APR_UINT64_T_FMT
" in revision %ld"),
- apr_off_t_toa(pool, offset),
+ offset,
rep->item_index, rep->revision);
Index: subversion/libsvn_fs_x/noderevs.c
===================================================================
--- subversion/libsvn_fs_x/noderevs.c (revision 1589948)
+++ subversion/libsvn_fs_x/noderevs.c (working copy)
@@ -436,6 +436,4 @@
return svn_error_createf(SVN_ERR_FS_CONTAINER_INDEX, NULL,
- apr_psprintf(pool,
- _("Node revision index %%%s"
- " exceeds container size %%d"),
- APR_SIZE_T_FMT),
+ _("Node revision index %" APR_SIZE_T_FMT
+ " exceeds container size %d"),
idx, container->noderevs->nelts);
]]]
The patch:
[[[
Index: tools/po/po-update.sh
===================================================================
--- tools/po/po-update.sh (revision 1589948)
+++ tools/po/po-update.sh (working copy)
@@ -49,2 +49,29 @@
+# Process the content of file "$1" into a new file named "$2".
+preprocess_file()
+{
+ sed < "$1" \
+ -e 's/APR_SSIZE_T_FMT/"ld"/g' \
+ -e 's/APR_SIZE_T_FMT/"lu"/g' \
+ -e 's/APR_OFF_T_FMT/"ld"/g' \
+ -e 's/APR_PID_T_FMT/"d"/g' \
+ -e 's/APR_INT64_T_FMT/PRId64/g' \
+ -e 's/APR_UINT64_T_FMT/PRIu64/g' \
+ -e 's/APR_UINT64_T_HEX_FMT/PRIx64/g' \
+ -e 's/SVN_FILESIZE_T_FMT/PRId64/g' \
+ > "$2"
+}
+
+# For each file name on standard input, process the file's content into a
+# new file and write the new file's name on standard output.
+preprocess_files()
+{
+ while read F; do
+ G="preprocessed/${F#../}"
+ mkdir -p $(dirname "$G")
+ preprocess_file "$F" "$G"
+ echo "$G"
+ done
+}
+
make_pot()
@@ -54,2 +82,3 @@
(cd $svn_base/subversion/po && \
+ rm -r preprocessed && \
find .. \
@@ -61,2 +90,3 @@
-name "svn_fs_util.h" -print | \
+ preprocess_files | \
$XGETTEXT --sort-by-file -k_ -kN_ -kQ_:1,2 -kSVN_ERRDEF:3 \
]]]
Thoughts?
- Julian
--
Join WANdisco's free daily demo sessions on Scaling Subversion for the
Enterprise
<http://www.wandisco.com/training/webinars>
