Re: Perform COPY FROM encoding conversions in larger chunks

On Mon, Feb 1, 2021 at 12:15 PM Heikki Linnakangas <hlinnaka(at)iki(dot)fi> wrote:
> Thanks. I fixed it slightly differently, and also changed LocalToUtf()
> to follow the same pattern, even though LocalToUtf() did not have the
> same bug.

Looks good to me.

> I added a bunch of tests for various built-in conversions.

Nice! I would like to have utf8 tests for every category of invalid byte
(overlong, surrogate, 5 bytes, etc), but it's not necessary for this patch.

> I spent some time refactoring and adding comments all around the patch,
> hopefully making it all more clear. One notable difference is that I
> renamed 'raw_buf' (which exists in master too) to 'input_buf', and
> renamed 'conversion_buf' to 'raw_buf'. I'm going to read through this
> patch again another day with fresh eyes, and also try to add some tests
> for the corner cases at buffer boundaries.

The comments and renaming are really helpful in understanding that file!

Although a new patch is likely forthcoming, I did take a brief look and
found the following:

In copyfromparse.c, this is now out of date:

* Read the next input line and stash it in line_buf, with conversion to
* server encoding.

One of your FIXME comments seems to allude to this, but if we really need a
difference here, maybe it should be explained:

+#define INPUT_BUF_SIZE 65536 /* we palloc INPUT_BUF_SIZE+1 bytes */

+#define RAW_BUF_SIZE 65536 /* allocated size of the buffer */

Lastly, it looks like pg_do_encoding_conversion_buf() ended up in 0003
accidentally?

--
John Naylor
EDB: http://www.enterprisedb.com