> -----Original Message-----
> From: Dmitry Kuzmenko [mailto:k...@ibase.ru] 
> Sent: Martes, 15 de Septiembre de 2015 7:47
> 
> default "no_rec_version" for read_committed also confuses people,
> because it "blocks reading of concurrent uncommitted changes".
> Jiri said that
> "that was Dmitry Kuzmenko's wild ride 3-4 years back."
> 
> and I was really mad by this. Because developers implementing drivers
> were not thinking about usability of isolation levels, and
> implementing "defaults" as is. And I said that "seems that they
> are not writing applications using their drivers".

Dmitry, I agree wholeheartedly with you.
If the path for newcomers is not easy, they go elsewhere.
Mind you, for years I had the idea that the default for read committed was
record version but since I've seen many documents about different
connectivity layers, I should have thought I read it in the IB manuals.
:-)


> BDE until version 4.1 had that read_committed
> behavior - blocking readers, mostly silent.
> I reported it to Borland, it was fixed in 4.1 and later.
> 
> Firebird ODBC had that behavior, until 1.5.2 version (IIRC)
> http://tracker.firebirdsql.org/browse/ODBC-120
> great thanks that it was fixed.

It's of no gain to preach from the roof with a set of theoretical
discourses. Make it easy for programmers that expect a typical behavior and
leave the more powerful options for the developers that care.


> Firebird .Net driver had (have?) this behavior
> http://tracker.firebirdsql.org/browse/DNET-337
> ? Jiri said that "it's up to developer that uses this driver".
> Can you imagine developer that starts to write first application with
> Firebird and knows ALL firebird transaction parameters, including
> defaults? Me - no.

Arrogance won't make us more popular nor it will increase the user base.
Take SqlServer for example, it has several quirks but it's easy to use with
the default options.


> Current defaults mostly the reason of Firebird being blamed, silently.
> People ties to work with FB, find strange behavior, then drops using
> it not asking anyone about what happens.

This is what any of us would do when needing to have a prototype in a rush:
use what works, no time to become an expert in some DB product. For example,
despite read committed being despised by the original creators of the
engine, it's useful in a traditional C/S application and the default mode
for FB (snapshot) is more appropriate for reports. Now that most of requests
are quick sequences of "connect, perform an action and disconnect" thanks to
the persistence frameworks, maybe this is not that important.

C.


------------------------------------------------------------------------------
Firebird-Devel mailing list, web interface at 
https://lists.sourceforge.net/lists/listinfo/firebird-devel

Reply via email to