Re: readdir is incorrectly implemented at Windows - Mailing list pgsql-hackers

From Andrew Dunstan
Subject Re: readdir is incorrectly implemented at Windows
Date
Msg-id 7b740522-3c48-1217-03f0-e2acf8f6b345@2ndQuadrant.com
Whole thread Raw
In response to readdir is incorrectly implemented at Windows  (Konstantin Knizhnik <k.knizhnik@postgrespro.ru>)
List pgsql-hackers
On 2/25/19 10:38 AM, Konstantin Knizhnik wrote:
> Hi hackers,
>
> Small issue with readir implementation for Windows.
> Right now it returns ENOENT in case of any error returned by
> FindFirstFile.
> So all places in Postgres where opendir/listdir are used will assume
> that directory is empty and
> do nothing without reporting any error.
> It is not so good if directory is actually not empty but there are not
> enough permissions for accessing the directory and FindFirstFile
> returns ERROR_ACCESS_DENIED:
>
> struct dirent *
> readdir(DIR *d)
> {
>     WIN32_FIND_DATA fd;
>
>     if (d->handle == INVALID_HANDLE_VALUE)
>     {
>         d->handle = FindFirstFile(d->dirname, &fd);
>         if (d->handle == INVALID_HANDLE_VALUE)
>         {
>             errno = ENOENT;
>             return NULL;
>         }
>     }
>
>
> Attached please find small patch fixing the problem.
>

Diagnosis seems correct. I wonder if this is responsible for some odd
things we've seen over time on Windows.

This reads a bit oddly:


             {
    -            errno = ENOENT;
    +            if (GetLastError() == ERROR_FILE_NOT_FOUND)
    +            {
    +                /* No more files, force errno=0 (unlike mingw) */
    +                errno = 0;
    +                return NULL;
    +            }
    +            _dosmaperr(GetLastError());
                 return NULL;
             }


Why not something like:


    if (GetLastError() == ERROR_FILE_NOT_FOUND)
        errno = 0;
    else
        _dosmaperr(GetLastError());
    return NULL;


cheers


andrew


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



pgsql-hackers by date:

Previous
From: Nikita Glukhov
Date:
Subject: Re: SQL/JSON: JSON_TABLE
Next
From: Andres Freund
Date:
Subject: Re: Refactoring the checkpointer's fsync request queue