Philip Martin wrote:
> URL: http://svn.apache.org/viewvc?rev=1431279&view=rev
> Log:
> Ensure wc.db has the permissions allowed by umask.
... in order to be consistent with the other files we write,
and to fix sharing a WC?
If this bug was breaking sharing a WC, should we back-port this fix?
> * subversion/libsvn_wc/wc_db_util.c
> (svn_wc__db_util_open_db): Ensure wc.db exists before it is opened by
> SQLite.
> +#ifndef WIN32
> + else
> + {
> + apr_file_t *f;
> +
> + /* A standard SQLite build creates a DB with mode 644 ^ !umask
Should the expression say "0644 & ~umask", using C syntax?
> + which means the file doesn't have group/world write access
> + even when umask allows it. By ensuring the file exists before
> + SQLite gets involved we give it the permissions allowed by
> + umask. */
It would be good to have a hint in the comment about *why*, even if just "for
consistency".
> + SVN_ERR(svn_io_file_open(&f, sdb_abspath,
> + (APR_READ | APR_WRITE | APR_CREATE),
> + APR_OS_DEFAULT, scratch_pool));
> + SVN_ERR(svn_io_file_close(f, scratch_pool));
> + }
> +#endif
>
> SVN_ERR(svn_sqlite__open(sdb, sdb_abspath, smode,
> my_statements ? my_statements : statements,
If we consider this to be a fix for the way SQLite behaves, do we want the same
throughout Subversion? Move the fix into svn_sqlite__open()? File a bug
report with SQLite project?
- Julian
