Обсуждение: pg_restore error message

On 2020-May-07, Euler Taveira wrote:

> While investigating a pg_restore error, I stumbled upon a message that is
> not so useful.
> 
> pg_restore: error: could not close data file: No such file or directory
> 
> Which file? File name should be printed too like in the error check for
> cfopen_read a few lines above.

Thanks for reporting.  Fix pushed to 9.5 and up.

-- 
Álvaro Herrera                https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services

On 2020-May-07, Ranier Vilela wrote:

> Can suggest improvements?
> 
> 1. free (398 line) must be pg_free(buf)';

Yeah, there's a lot of frontend code that uses free() instead of
pg_free().  There are too many of these that worrying about a single one
would not improve things much.  I guess we could convert them all, but I
don't see much point.

> 2. %m, is a format to parameter, right?
>     But what parameter? Both fatal call, do not pass this parameter, or is
> it implied?

%m is an implied "strerror(errno)", implemented by our snprintf
replacement.

-- 
Álvaro Herrera                https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services

On Fri, May 08, 2020 at 07:45:16PM -0400, Alvaro Herrera wrote:
> Yeah, there's a lot of frontend code that uses free() instead of
> pg_free().  There are too many of these that worrying about a single one
> would not improve things much.  I guess we could convert them all, but I
> don't see much point.

Doing a hard switch would have the disadvantage to create more
problems when back-patching.  Even if such conflicts would be I guess
simple enough to address, that's less to worry about.  I think however
that there is a point in switching to a more PG-like API if reworking
an area of the code for a new feature or a refactoring, but this is a
case-by-case judgement usually.

>> 2. %m, is a format to parameter, right?
>>     But what parameter? Both fatal call, do not pass this parameter, or is
>> it implied?
>
> %m is an implied "strerror(errno)", implemented by our snprintf
> replacement.

Originally, %m is a glibc extension, which has been added recently in
our port in src/port/snprintf.c as of d6c55de.
--
Michael