Re: Patch for collation using ICU - Mailing list pgsql-hackers

From Magnus Hagander
Subject Re: Patch for collation using ICU
Date
Msg-id 6BCB9D8A16AC4241919521715F4D8BCE6C7430@algol.sollentuna.se
Whole thread Raw
In response to Patch for collation using ICU  (Palle Girgensohn <girgen@pingpong.net>)
Responses Re: Patch for collation using ICU  (Andrew Dunstan <andrew@dunslane.net>)
List pgsql-hackers
>> The 3.2 vs 2.8 business is disturbing also; specifically, I
>> don't think we get to require 3.2 on a platform where 2.8 is
>> installed.
>
>There seems to be nothing in the ICU licence that would prevent us from
>bundling it.
>This would solve both the 3.2 vs 2.8 problems, and would remove the
>'dependency'.
>
>> People just aren't going to hold still for that, even
>> assuming that ICU supports installing both versions at once,
>> which isn't clear to me at the moment ...
>
>There's no problems with having both installed.

... unless you're on win32, it seems.  For some reason, they name their
libs with the version on unix (libicu18n.so.32), but not on win32 where
they all have the same name. And they don't stuff versioning information
in the DLL files.
That can be lived with as long as libpq doesn't depend on it, though -
you can just stick the DLL in the same directory as the EXE, which is
also what the ICU people recommend in their docs. Unnecessarily ugly,
but it works.


>I did that on debian to get the patch going.
>Tho, bundling it seems cleaner to me.

The source for ICU 3.2 is 9.8Mb in .tar.gz. PostgreSQL 8.0.2 is 13.2.
That means the size of the distribution would almost *double* if we
bundled ICU.

It's probably fine bundling it in the binary distributions (at least
we'd probably do it on win32, since not many ppl will have it already
there), but bundling the source seems a bit excessive to me.

//Magnus


pgsql-hackers by date:

Previous
From: "Magnus Hagander"
Date:
Subject: Re: Patch for collation using ICU
Next
From: Tatsuo Ishii
Date:
Subject: why two WRITE_NODE_FIELD(whereCluase)?