Pike-devel April 2011

pike-devel@lists.lysator.liu.se

6 participants
11 discussions

Sql.pgsql behavior in case of errors
by aldem-pikedev＠aldem.net 21 Aug '14

21 Aug '14

Hi everyone, I am preparing to use Postgres in my app, found the following strange behavior: ---snip--- Sql.Sql db = Sql.Sql("pgsql://localhost", "test", "test", "test"); db->query("BEGIN"); db->query("INSERT INTO requests (SvTrId) VALUES ('lala')"); catch(db->query("INSERT INTO requests (SvTrId) VALUES ('lala')")); db->query("INSERT INTO requests (SvTrId) VALUES ('juju')"); db->query("ROLLBACK"); ---snip--- Database is opened, row with 'lala' is inserted, second query (obviously) fails as SvTrId is the primary key, but row with 'juju' is inserted *and* appears in the database - i.e. the transaction is *committed*! Studying source code show, that once error is encounter, session is reset, i.e. running transaction is rolled back, everything is closed, temporary tables are deleted and session variables are reset as well, and after that, the next statement(s) will be committed immediately, as there is no active transaction. In my opinion, this is not exactly correct behavior when statement fails, as session still may be needed (especially temporary tables, open cursors etc). Im some cases, this is perfectly normal when SQL statement fails, especially in SERIALIZED transaction isolation level, and it would be more practical to leave the decision as of what to do to the application. Yes, this is true that it is impossible to continue executing statements in failed transaction, but temporary tables etc. may still be needed. Also, silent reconnection is not a good idea too - when there are many transactional statements are executed - it would be nice to have an option to modify this behavior. Best regards, /Alexander

5 6

Re: Calendar.Month / Calendar.Hour
by Marc Dirix 23 Aug '11

23 Aug '11

On 04/12/2011 11:05 AM, Mirar @ Pike developers forum wrote: > That's an old bug that I thought would be fixed by now. Pike optimizes > the + call and calls `+ with more than one argument. The Calendar > module ignores all other arguments than the first. > > (I think it's that one anyway.) Ok, then I have to wait for the new debian package (still on 7.8.352)

8 12

foreign_ident frustration
by Martin Stjernholm, Roxen IS ＠ Pike developers forum 17 Apr '11

17 Apr '11

I'm getting quite fed up fiddling with those foreign_ident markers. A serious problem is that the changes in .gitattributes makes it cumbersome to cherry pick patches between branches and reorder commits (before push, of course). The .gitattributes diffs also preclude the possibility for git to recognize the same patch on the different branches. Another annoyance is that the commit checks get in the way: I never remember to check for foreign_idents before committing, so the commit get interrupted and I have to salvage the commit message (which in the case of using MaGIT in Emacs is definitely lost). Often it's also necessary to reset the file to get the $Id$ expansion right, which means I have to restage the commit with chunks and everything. On top of that there's various strange behavior in git with "spook differences" when the $Id$ handling changes. E.g. today I got into a state with a file where git reset/checkout resulted in it being considered changed in the $Id$ expansion right from the start. Killing the chunk didn't work, and it simply wasn't possible to get the working tree clean with the intended git tools. Things like this are bugs in git, but be that as it may, they add to the frustration. This foreign_ident system is very proper and correct, but it's frankly a big PITA, and I don't see any benefits with it that even comes close to make it worth the hassle. Keeping the cvs idents around in the historical versions is one thing, but replacing them gradually instead of wholesale is just a little more neat, perhaps. The whole thing boils down to avoiding just a single extra commit. I don't think it's a passing phase either - there are a lot of files, and it'll take a long time for them all to change naturally (it'll probably never happen). Also, the problem with backporting patches will only get worse as the current branch loses more foreign_idents compared to the old ones. So is there a compelling reason to keep those foreign_ident around, or would it be ok to drop them all in one go?

3 5

Re: Calendar.Month / Calendar.Hour
by Marc Dirix 12 Apr '11

12 Apr '11

On 12-04-11 11:20, Mirar @ Pike developers forum wrote: > Or just do ->day(x)->minute(y) or so. :) Yes, but Calendar.Day()->day(-2)->minute(460); for example doesn't work. (At least in 7.8.352): > Calendar.Day()->day(-2)->minute(460); Not in range (Day 1..1 exist). /usr/lib/pike7.8/modules/Calendar.pmod/YMD.pike:818: Day(Tue 12 Apr 2011)->day(0) HilfeInput:1: HilfeInput()->___HilfeWrapper()

2 1

Calendar.Month / Calendar.Hour
by Marc Dirix 12 Apr '11

12 Apr '11

Hi, Pike 7.8.352 (Debian) In hilfe: > Calendar.now(); (21) Result: Fraction(Thu 7 Apr 2011 12:26:15.216691 CEST) > Calendar.Hour()+Calendar.Minute(); (22) Result: Minute(Thu 7 Apr 2011 12:01 CEST - 13:01 CEST) > Calendar.Month()+Calendar.Minute(); (23) Result: Minute(Fri 1 Apr 2011 0:01 CEST - Sun 1 May 2011 0:01 CEST) So Hour is calculated from the beginning of the last hour + 1 minute, and Month calculates starting of the month until next month + 1 minute. I expected the behaviour to be the same as with Calendar.Hour()? Best regards, Marc Dirix

4 6

Re: Calendar.Month / Calendar.Hour
by Marc Dirix 09 Apr '11

09 Apr '11

Thank you for thinking with me! I know the number of minutes, I just have to convert them to a date so I can test if current-time >= schedule-time. I think I need something along the lines of: next = Calendar.Day()->beginning() + Calendar.Minute()*x; or for tomorrow: next = Calendar.Day()->end() + Calendar.Minute()*x; /Marc

3 5

call_outs
by Arne Goedeke 08 Apr '11

08 Apr '11

I was about to change backend.cmod to use power of two hashtables, aswell. Its using hash_svalue anyhow, and it seems a little unnecessary to use mod prime hashtables there. Anyhow, while doing so and putting some benchmarks together I noticed that the code in backend.cmod:990 ff assumes that is_eq has side effets. Because of that it does alot of extra work. Its especially bad if one has alot of callouts to the same function since the buckets get very large. If a call to remove_call_out/call_out is the only thing that could change the hash tables, we might aswell do some kind of recursion detection and error instead (in case someone does call_outs from `==). Any thoughts? arne

4 11

Re: Calendar.Month / Calendar.Hour
by Marc Dirix 07 Apr '11

07 Apr '11

I need to calculate a time based on the number of minutes since the start of the month, or since the start of the day. Calendar.Day()+Calendar.Minute()*x always worked fine, but Calendar.Month() + Calendar.Minute()* doesn't. But probably I have to revise the code to use beginning() everywhere, also for Calendar.Day(). /Marc

2 1

Re: Calendar.Month / Calendar.Hour
by Marc Dirix 07 Apr '11

07 Apr '11

Ow, my mistake. You're right.

1 0

forced cleanup-on-exit
by Arne Goedeke 02 Apr '11

02 Apr '11

I am currently writing a cmod for some obscure piece of hardware. Its driver tends to generate kernel panics if some deinitialization does not happen. It would therefore be nice if it was somehow possible to force the destruction of certain objects on exit, even without --with-cleanup-on-exit. Is there a possibility for that? And no, I cannot fix the kernel driver ;-)

3 4

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

2009

2008

2007

2006

2005

2004

2003

2002

Pike-devel April 2011