Ticket #122: modeldiscuss.txt

File modeldiscuss.txt, 31.9 kB (added by hugo <gb@bofh.ms>, 3 years ago)
IRC discussion about the model syntax

Line
1	* Now talking on #django
2	* Topic for #django is: http://www.djangoproject.com/ \|\| Django is a high-level Python Web framework that encourages rapid development and clean, pragmatic design.\|\| Logs: http://loglibrary.com/show_page/latest/179
3	* Topic for #django set by JZ_ at Sat Jul 16 12:30:01 2005
4	hugo- hi
5	mmarshall hi hugo-
6	* adrian_h is back.
7	adrian_h back
8	mmarshall Ok everyone, we can start talking again ;)
9	rmunn adrian_h: You said you're split on #122. What are your main objections?
10	* ljoramo has quit ()
11	* adrian_h gathers thoughts
12	rmunn Take your time. :-)
13	adrian_h I don't like the fact that ForeignKeys are name-less attributes of the class (yes, I was the one who suggested that behavior in the first place)
14	rmunn That would be a little surprising for newbies, yeah.
15	adrian_h Before we open-sourced Django, the field name was required. Foreign keys used to look like this:
16	adrian_h meta.IntegerField('field_name', 'verbose name', rel=meta.ManyToOne(Poll))
17	adrian_h So I shortened that to "ForeignKey(Poll)" in preparation for open-sourcing
18	* Netminder likes the current state of fields FWIW.
19	adrian_h But the one thing that was lost in that was that now you have to "guess" what the field names are, for use in meta.admin.fields
20	adrian_h I talked about this with MikeLambert last night
21	rmunn I personally would have little objection to making ForeignKeys require a name again...
22	hugo- me too
23	rmunn It's orthogonal to the question of the SQLObject-like syntax, IMHO.
24	rmunn Actually, requiring ForeignKeys (and ManyToManyFields) to have a name would get rid of one of the more "magic" parts of the #122 patch.
25	adrian_h As I see it, the issues are related, because one solution can solve both problems
26	adrian_h Yeah, exactly
27	rmunn They're related, but independent: it would be quite feasible to implement one, or the other, or both, or neither.
28	rmunn One of those four is probably optimum, we just need to figure out which.
29	rmunn Another possibility: in mmarshall's email to the dev list, he mentioned writing a wrapper class for the new behavior.
30	rmunn I would suggest making the new behavior the default, but keeping a wrapper class ("meta.OldModel" perhaps) around to give the old "fields tuple" behavior.
31	adrian_h Yeah, that's possible...But there should be One Real Way of doing it
32	adrian_h Yes, we've discussed meta.OldModel (with exactly that name)
33	rmunn The documentation would mention the "fieldname = CharField()" syntax only, and the BackwardsIncompatibilities page would say "As a transition measure, use meta.OldModel until you've converted all your models".
34	mmarshall rmunn: That was what I tried to do at first, but there were some problems.
35	rmunn mmarshall: What problems did you run into?
36	adrian_h IIRC, the problem lies in the fact that django/models/__init__.py can't be duplicated
37	adrian_h i.e., we can't have multiple copies of it
38	mmarshall For one thing, the changes were not local meta.Model; they also required switching the first two arguments to the Field classes.
39	rmunn Ah yes, so that verbose_name could become the first positional arg.
40	mmarshall So I tried duplicating meta itself, resulting in a new can of worms.
41	mmarshall rmunn: Yes.
42	mmarshall At first, I had a hack to get around this, but it was messy, and prone to confuse people.
43	rmunn So if this is going to happen, it sounds like the simplest way is to just bite the bullet and say, "OK folks, revision ### is going to break all your models. Here's how to convert to the new syntax."
44	adrian_h Yeah...And, realistically, having to change models to use the "fieldname = class" syntax is only slightly more work than changing models to use "meta.OldModel"
45	rmunn How would you feel about the "fieldname = class" syntax if the decision was made that ForeignKeys must always be given a name?
46	mmarshall Yeah, it's really not that big of a deal. Although, after converting all of the models in the tests and the django internals, as well as my own, I got a little sick of it.
47	* threeve has quit ()
48	rmunn Any other major objections?
49	rmunn That was to adrian_h, of course, not mmarshall. :-)
50	rmunn mmarshall: But it's a one-time price.
51	adrian_h My only other major objection is having to ask everybody to change model syntax
52	adrian_h Which isn't a huge problem, being that we haven't guaranteed backwards-compatibility
53	rmunn Well, there's no getting around that one.
54	rmunn It'll mostly depend on whether people want the new syntax or not.
55	adrian_h And yesterday's session update was backwards-incompatible, and people didn't complain too loudly...
56	Netminder adrian_h: very few people had gotten to sessions yet
57	rmunn If there are loud boos and very few cheers, you could always revert the patch...
58	mmarshall Yeah, but that wasn't as big of deal to change.
59	rmunn I, for one, would be cheering loudly. :-)
60	Netminder changing the model syntax is pretty big.
61	adrian_h Netminder: Yes, but the session change broke admin installations, because the admin now depends on sessions
62	rmunn Yeah, it is, which is why I'm trying to start discussion on the dev list.
63	adrian_h People haven't really spoken up about the model syntax, other than rmunn and mmarshall
64	mmarshall I have had a number of people mention it to me.
65	rmunn manuzhai was saying earlier in here that he planned to write a post to the dev list.
66	* Tybstar (n=tgerla@cpe-069-134-162-085.nc.res.rr.com) has joined #django
67	Netminder adrian_h: I'd probably be mildly against it, as I don't feel a "big gain" in the new syntax vs. the current and it'd be a good bit of work and relearnin'
68	adrian_h How does manuzhai feel about it?
69	rmunn loglibrary's not working, so I'll just copy and paste:
70	rmunn manuzhai: "will send a me too to the #122 thread once I get an email from it"
71	adrian_h Netminder: all right
72	* Tybstar returns
73	adrian_h Tybstar: How do you feel about the proposed model syntax change to use "fieldname = FieldClass"?
74	mmarshall When I came in earlier, he mentioned that #122 was 'shot down', and frowned.
75	rmunn manuzhai: "I for one like 'your' syntax better" (to mmarshall)
76	Netminder adrian_h: I heard some rumbles on the ticket about issues with the db layer. Is that a show stopper or just something to get around?
77	* liquidx (n=liquidx@87.240.134.143) has joined #django
78	hugo- I don't think that it really is much change in the syntax. Actually that's the reason why I am a bit split on #122, too - I don't see much gain for the work
79	Tybstar adrian_h: i haven't really followed the discussion, unfortunately. i liked it on first glance, but i've really gotten attached to the current syntax.
80	Netminder yes, that's my thought too hugo-
81	hugo- ok, sure, the new syntax would be nicer, but is it really that big a problem to use a list of fields instead of a list of attribute assignments?
82	Tybstar it's not a huge difference either way...
83	rmunn It's not a huge gain once you've gotten used to the fields = (...) syntax. I see it more as a big gain for newbies.
84	hugo- is it really?
85	Netminder It's just... different.
86	adrian_h Netminder: There shouldn't be any issues with the DB layer with regard to the possible syntax change -- the syntax is abstracted
87	Netminder adrian_h: roger that.
88	hugo- if I think about what questions arise here in the channel, I don't think that _that_ part of Django really produces that much questions
89	rmunn Ah good, some real discussion at last. :-)
90	adrian_h Great point, hugo-
91	hugo- there are other things like the settings file an dstuff like that :-)
92	Tybstar the model fields syntax isn't bad. the admin = syntax can be pretty hairy.
93	Tybstar i've seen lots of people have trouble wtih the (foo) != (foo,) thing in the admin = code
94	rmunn I've seen that too.
95	hugo- Tybstar: yeah, that's one thing I allways want to hit Guido over the head because of ;-)
96	Netminder the (foo,) syntax becomes second nature after a messup or two
97	rmunn I'd suggest a general rule-of-thumb: if something is supposed to be a list or tuple, and you get a string, automatically convert it.
98	adrian_h The one real problem I have with the model syntax is that it's not immediately apparent what the field name is for a ForeignKey and ManyToManyField, so people don't know what to use in admin.fields
99	Tybstar adrian_h: indeed
100	Tybstar takes some guesswork
101	hugo- Tybstar: especially since python will happily iterate over chars in a string if it expects a list :-)
102	Tybstar yes
103	adrian_h We can fix that with the model validator, with a message that says "That's an invalid field name. Here are the valid field names for this model", but that's still one step too much
104	Tybstar the pitfalls of a dynamically typed langage. :)
105	hugo- adrian_h: I would applaud a move that ForeignKey and friends need a obligatory first parameter that is the field name like other field types
106	rmunn That's what ticket #244 was about.
107	hugo- adrian_h: especially since that would make writing models for existing databases much more obvious
108	hugo- adrian_h: especially since the "name=" option for ForeignKey isn't (or at least wasn't) documented in the model reference :-)
109	rmunn Wow, discussion moved on while I was searching for that ticket.
110	rmunn I meant that http://code.djangoproject.com/ticket/244 was about being nicer to newbies.
111	hugo- rmunn: you need to configure --with-threads ;-)
112	rmunn Heh. I wish IRC had that sometimes... :-)
113	adrian_h :)
114	adrian_h hugo-: Would you applaud the move if it were in "fieldname = FieldClass" syntax?
115	Netminder irc --with-productivity would be nice
116	mmarshall Netminder: LOL
117	rmunn Netminder: :-D
118	rmunn Okay, let me see who seems to like or dislike the "fieldname = FieldClass" syntax.
119	rmunn I like it.
120	rmunn mmarhsall likes it.
121	rmunn adrian_h is split but leaning against (?)
122	rmunn Netminder isn't too enthusiastic.
123	hugo- adrian_h: I wouldn't mind if it was, as I don't have much models to change - but I still wouldn't see the reason to switch that part of the syntax. But I wouldn't actively object.
124	rmunn Tybstar liked it at first but is pretty indifferent.
125	rmunn hugo- doesn't much care.
126	mmarshall garthk likes it, manuzhai likes it
127	hugo- rmunn: doesn't much care with added a bit of puzzlement, please ;-)
128	rmunn Anyone object to how I've described them?
129	adrian_h I'm positively for requiring a name for ForeignKey and ManyToManyField, to make that explicit -- which is a related issue
130	hugo- adrian_h: that one I am quite positive about, too
131	rmunn adrian_h: I'm +1 on requiring a name as well.
132	Netminder rmunn: that's about right. +1 on the requirement
133	mmarshall +1 on requiring a name here
134	Tybstar +1
135	adrian_h Wow, no opposition to the name requirement -- very cool
136	rfc1149 count me in :) +1 on requiring a name
137	Tybstar sounds unaniniminninnious
138	mmarshall I normally specify a name anyway :P
139	adrian_h rfc1149: What do you think of the "fieldname=FieldClass" syntax?
140	adrian_h As an aside, can anybody on Windows confirm this bug? http://code.djangoproject.com/ticket/349
141	rfc1149 i quite like the fieldname=FieldClass idea, but am a bit concerned that it will bite later. using fields= leaves a lot of namespace for later use.
142	adrian_h rfc1149: The idea is that other (non-field) data would go in an inner class called "Meta"
143	rfc1149 btw, i also would like to interpret "foo" as ("foo",) in settings where it makes sense
144	rfc1149 oh, that would be very nice!
145	rfc1149 (the inner Meta class, that is)
146	rmunn What I see is support for fieldname=FieldClass ranging from mild to enthusiastic, but I don't see any opposition stronger than mild.
147	rmunn Is there anyone who would positively scream if that was implemented?
148	rmunn Remember, it would be a backwards-incompatible change to the model syntax, so it would involve work (albeit a one-time effort) on your part to change your models.
149	rfc1149 i do not yet have significant amounts of existing code, but i would not like it if an unthinking svn update would break all my code
150	* Netminder remains mildly against it.
151	rfc1149 but i guess that is unavoidable :)
152	adrian_h Here's an example of the new syntax, so people have something to look at: http://django.pastebin.com/339441
153	Netminder I'd kinda grumble a bit while rewriting my todo app :)
154	rfc1149 and it's still not yet released :)
155	* Boffbowsh (i=Boffbows@host-85-236-105-19.multiplaydsl.co.uk) has joined #django
156	rfc1149 the field=FieldClass syntax really looks clean and natural
157	rmunn Boffbowsh: Have you looked at http://code.djangoproject.com/ticket/122 at all?
158	rfc1149 and friendly to newbies.. i'm liking it more and more.
159	hugo- I'd prefer it if the inner class wasn't tacked on by magic but instead just be added via a __meta__ = something attribute
160	Netminder yeah, it's kinda nice. My only objection would be the one time fee that we'd have to pay, but I think it would be worth it in the long run.
161	Boffbowsh yeah, i've taken a look, if thats'the fieldname = fielddef thing
162	Boffbowsh i agree it would be nice
163	rmunn I'm trying to get a feel for whether people would really like that syntax, really hate it, or be indifferent.
164	Netminder yeah, I'm not keen on the magic
165	hugo- field names with __ in front and back could be defined as "internal use" - that would match with python thinking
166	Boffbowsh but not at the expense of ease of maintenance of django, imo
167	Netminder +1 on hugo's line of thought
168	Boffbowsh yar
169	rfc1149 Boffbowsh: that's a good point
170	rmunn hugo-: Interesting idea.
171	rfc1149 but i have not really heard adrian_h complain about it
172	adrian_h If we use fieldname=fielddef, the maintenance will be about the same as it is now
173	rmunn Perhaps __meta__ = dict(admin=Admin(), foo=Bar(), ...)
174	rfc1149 see :)
175	Boffbowsh i don't mind having a slightly odd syntax (which isn't really that odd at all, there's just a slightly nicer way) if it means that the rest of the project is stable and easy to add new stuff to
176	adrian_h But the last patch to #122 (which allows model attributes without names) would be a headache to maintain
177	Boffbowsh not read it through thoruoghly as i was at work when it first cropped up
178	rfc1149 i personally prefer Meta: ... above __meta__ = dict( ... )
179	adrian_h Using a class for Meta would be better, because you wouldn't have to use commas between attributes
180	Netminder hmm yeah
181	adrian_h Models themselves used to be dictionaries, a long time ago
182	Boffbowsh are there any benefits other than it looking nicer?
183	hugo- adrian_h: yeah, but the problem is with the way how you find out in your Poll class that there is an inner class Meta
184	mmarshall adrian_h: how would it be hard to maintain?
185	rmunn There seems to be pretty much unanimous consensus to require a name for ForeignKey and ManyToManyFields, so the headache-to-maintain patch can probably go away.
186	Netminder if it looks prettier but is painful to maintain, that's no good
187	hugo- there is no obvious connection between those two besides the scope - the internally needed assignments are done by magic
188	adrian_h mmarshall: I was referring to the magic no-name-for-ForeignKeys thing, but we've agreed to scrap that.
189	* rfc1149 goes to bed but looks forward to reading the rest of the discussion later
190	* Dagur (n=dagur@dsl-196-5.hive.is) has joined #django
191	rmunn rfc1149: Don't log out -- loglibrary is down.
192	hugo- and more specificially: there is no prior art in Python libraries for such magic that connects stuff implicitely just because of the scope they are in
193	adrian_h hugo-: You're talking about the internal "class Meta", right?
194	hugo- adrian_h: yep
195	hugo- I agree that it's cleaner syntax, though. <sigh> not an easy choice
196	hugo- the hacker in me would go the inner-class-way, the maintainer might shy away from it :-)
197	Netminder I'm still on the fence, but think maintainability is a key thing to think about
198	Netminder the syntax is quite nice though :)
199	adrian_h The maintenance for the inner class is simple -- internally (inside the metaclass call), it's just interpreted as another attribute on the object
200	rmunn I'm looking at things more from a usability perspective than a maintainability perspective, so that colors my perception.
201	rmunn Most of the tickets I file tend to be useability issues, if you've noticed.
202	adrian_h It's even probably a bit cleaner than the current internals
203	adrian_h rmunn: Yes, and they're much appreciated :)
204	Dagur Could someone tell me what this error means:
205	Dagur ContentTypeDoesNotExist: ContentType does not exist for {'package__label__exact': 'jenga', 'python_module_name__exact': 'jengacats'}
206	Dagur I get it when I try to submit stuff in the admin panel
207	rmunn Dagur: Have you changed your model recently?
208	Dagur yeah
209	hugo- adrian_h: the nice thing about the current stuff is: it's all python data. The new stuff is more python code objects. It's easy to programmatically generate python data if there is much repetitive stuff in it - but it's much harder (although not impossible thanks to pythons structure) to dynamically construct code objets
210	adrian_h Dagur: run "django-admin.py sqlall jenga" and execute the "INSERT INTO content_types ... jenga" line in your database.
211	rmunn Did you update the database after changing the model?
212	Dagur it's a model I created for an existing database
213	Dagur I didn't try that
214	Dagur will do
215	rmunn Dagur: Then you'll need to run sqlall, like adrian_h said.
216	rmunn But you'll also need to add all the auth_* and core_* tables.
217	hugo- adrian_h: and since I am in a line of business where there are often very large repetitive strikes of code and field definitions, I am much in favor of easy generating things programmatically
218	rmunn So that the INSERT INTO statements have somewhere to put the data. :-)
219	adrian_h Dagur: After doing "django-admin.py init", just do "django-admin.py sqlinitialdata jenga"
220	hugo- adrian_h: so in that aspect the maintainer in me shys away from solutions that leave the data-path and go the code path, even though the hacker in me prefers to hack on the latter stuff
221	hugo- usually the maintainer in me kicks the butt of the hacker in me ...
222	Dagur ok i'm confused now
223	rmunn hugo-: I think, though I'm not sure, that mmarshall's patch contructs a "fields" tuple internally.
224	rmunn Dagur: Listen to adran_h, not me. He knows what he's talking about.
225	Dagur hehe ok
226	* rheinbot has quit (Remote closed the connection)
227	* Tybstar needs a viewsvn for his project
228	adrian_h Dagur: Run the command "django-admin.py init" if you haven't already done that. That creates Django's internal database tables.
229	Netminder Tybstar: see the one that slashzero put django_ajax on, it's free to all
230	* Tybstar ponders writing a django-based viewsvn
231	adrian_h Dagur: After you've done that, run "django-admin.py sqlinitialdata jenga" and execute that SQL in your database.
232	adrian_h Dagur: That's it.
233	hugo- Tybstar: use trac
234	* rheinbot (n=supybot@maxwell.servers.ljworld.com) has joined #django
235	rmunn hugo-: Trac's internal Subversion viewer annoys me a little.
236	Tybstar hugo-: looks kinda heavyweight
237	rmunn It doesn't let me see what the root of the repository is.
238	rmunn I have to guess what to put after "svn checkout".
239	hugo- Tybstar: it isn't - I use it for most of my repositories nowadays, it's easy to set up and run
240	hugo- rmunn: hey, there is a wiki in it, just add a description of the checkout path to the wiki :-)
241	* Notify: jacobkm is offline (kornbluth.freenode.net).
242	Dagur woo! it worked
243	Dagur thanks
244	adrian_h Dagur: No problem! Thanks for using Django
245	Netminder Dagur: drive through :)
246	Dagur ^_^
247	* Tybstar can't make viewsvn work
248	Dagur I have another error for you guys :P
249	adrian_h Dagur: Go ahead
250	Dagur TypeError: got unexpected keyword argument 'id__iexactjengausers'
251	rmunn hugo-: Not every project that uses Trac seems to do that, though. That's my main gripe. It's a pretty small one, as gripes go.
252	Dagur the problem is that the id fields are called ID (capitalized)
253	adrian_h Dagur: Where did that error happen?
254	Dagur i was submitting in the admin panel
255	* GvaderTH (i=grzegorz@mokotow.com) has left #django
256	adrian_h Dagur: Can you paste the full traceback to django.pastebin.com?
257	Dagur http://django.pastebin.com/339455
258	adrian_h Thanks...I'm looking now
259	Dagur it's the id of a foreignkey table
260	Dagur not the table I'm updating
261	* wilsonian has quit ()
262	Dagur oops
263	Dagur id__iexactjengausers <-- the jengausers part is something I added
264	Dagur it's supposed to say id__iexact
265	adrian_h OK, I was wondering about that :)
266	Dagur sorry, I was trying to figure out the table name
267	adrian_h No problem
268	mmarshall hugo-: My patch simply itenerates over all of the attributes. If the attribute holds an instance of Field, it assigns the name (or rel_name, for relationship fields) from the attribute name, and adds it to a 'fields' list. This list is then used just the same as the fields tuple is currently.
269	adrian_h mmarshall: Ideally it would just loop over all the attributes except the one (if given) named "Meta"
270	rmunn Dagur: Is that "iexact" a typo that's supposed to be "exact", or did you really mean "iexact"?
271	mmarshall adrian_h: Actually, that's what it does.
272	* GvaderTH (n=gregtech@mokotow.com) has joined #django
273	adrian_h mmarshall: Ah, cool
274	adrian_h rmunn: That iexact is within Django...I'm fixing now.
275	hugo- adrian_h: and so I can't name a field Meta? yuck.
276	mmarshall Also, if it comes accross an attribute that is neither a Field nor a function, it raises a helpful error.
277	Dagur rmunn: it's the django code that does that
278	Dagur I was just submitting a form
279	mmarshall hugo-: Originally I supported doing that, by optionally placing all fields in a subclass named 'Fields', but Adrian didn't like that.
280	adrian_h Dagur: "svn update" your code, and let me know whether the problem is fixed.
281	Dagur ok
282	hugo- hmm. the more I think it through, the more I would tend to -1 on the new field syntax
283	mmarshall hugo-: actually, you can still do that, just by giving the 'name' argument.
284	Dagur worked!
285	adrian_h Dagur: Thanks for pointing out that bug!
286	adrian_h Dagur: Sorry, I should've credited you in the commit message
287	Dagur np, my pleasure!
288	adrian_h hugo-: One solution to that might be to require that "class Meta" is a subclass of "meta.Meta" or something like that
289	adrian_h "class Meta(meta.Meta)" -- hehe
290	rmunn Ew, that's a little overboard.
291	rmunn What about "class __meta__: ..."
292	adrian_h Is that possible?
293	hugo- sure
294	rmunn I don't see why not, it's just another name in the namespace.
295	adrian_h Very cool -- I just tried it
296	rmunn Let me check and make sure.
297	* hugo- has a class named __proxy__ in one of his source files :-)
298	rmunn Oh, nevermind then.
299	adrian_h Now that's much better, namespace-wise
300	rmunn Anyone who's using double-underscores at the start and end of their field names is being unpythonic, IMHO.
301	adrian_h But it introduces more double underscores :-/
302	Tybstar i'd suggest against __meta__
303	rmunn Yeah, that's the downside.
304	Tybstar it implies python builtin magic
305	hugo- rmunn: ugh - python uses loads of __xxxx__ stuff - is it unpythonic? :-)
306	Tybstar _Meta would be better
307	Netminder eew
308	rmunn hugo-: Fields are public things, __xxxx__ is for private stuff.
309	Netminder I'd take __meta__ before _Meta
310	hugo- and there are special provisions in python with regard to __xxxxx fields - they are hidden in some aspects
311	adrian_h hugo-: That's actually an advantage in this case
312	hugo- rmunn: the meta stuff _is_ private to the model, not public :-)
313	adrian_h What about "class DjangoMeta"?
314	hugo- but the reason why I dislike the new stuff is that it doesn't allow me to do things like fields = tuple([IntegerField(name, default=0) for name in list_of_fieldnames]) or something similar as easily
315	adrian_h That's much less likely to be used as a field name
316	Tybstar Netminder: i don't necessarily like _Meta better than __meta__, but i'm afraid of collidiing with python internals someday
317	rmunn What I meant was that "class __meta__" won't interfere with anyone's DB field name, since you should be using double-underscores for DB fields (which are more "public" attributes).
318	adrian_h hugo-: Interesting point, but, realistically, why would you want to do that, other than laziness? Models don't change often enough for there to be a need for them to be dynamic like that.
319	hugo- sure, if I know about the inner workings I can hack around it - but I can't just take the public visible interface of the model and apply python code there
320	hugo- adrian_h: laziness is my second name ;-)
321	adrian_h hehe
322	hugo- adrian_h: and for a more serious reason: I am working in legacy apps areas where you get _large_ structures of many very similar fields
323	hugo- if you have to write models with 100 or more fields and most of them are booleans or moneytypes or some choice of a handfull of possibilities, you crave for automatted solutions
324	Tybstar my cat just caught a blue jay.
325	adrian_h hugo-: "print ','.join(["IntegerField(%s, default=0)" % name for name in list_of_fieldnames])"
326	hugo- adrian_h: nooo - I hate generated code, programmatic solutions should just manipulate data
327	hugo- so to do the same with the new stuff I would write a loop that does a load of setattr(...)
328	hugo- it's not impossible, it's just not as easy and simple as with data structures
329	adrian_h hugo-: setattr wouldn't even work, because the metaclass runs when the class is defined
330	hugo- ah, yes, forgotten that one
331	adrian_h I must say, the new syntax is sexy
332	rmunn I suppose you could construct a dict and feed it to meta.Model.__new__(cls, bases, attrs). Ew.
333	rmunn I don't even know if that would work.
334	hugo- rmunn: yep, with metaclasses that would be a - rather ugly, I must say - option
335	adrian_h Or dynamically construct a class and dynamically set its parent class after it's been constructed?
336	hugo- sure, you can allways call the metaclass manually and construct the class instance by using the metaclass. It's a bit hacky and doesn't sound like really maintainable code, though ;-)
337	rmunn adrian_h: The metaclass magic happens before the class object is created, so I don't think that would work.
338	* manuzhai (n=manuzhai@i249181.upc-i.chello.nl) has joined #django
339	rmunn hugo-: You'd have to really deeply grok metaclasses to do that.
340	manuzhai soooo, is this were the #122 discussion is taking place?
341	adrian_h hugo-: What about a helper function for cases like yours? meta.create_model_from_dynamic_field_list(), except with a better name
342	rmunn hugo-: In other words, don't try it unless you're Dutch. :-)
343	adrian_h manuzhai: Yes, indeed
344	rmunn manuzhai: Yes.
345	hugo- rmunn: the main problem is that stuff like that would be code that only I would maintain because my coworkers _never_ will grok it ;-)
346	manuzhai rmunn: hey, don't fuck with the Dutch!!!
347	rmunn manuzhai: No no, it's a compliment.
348	manuzhai ok
349	hugo- rmunn: hey, I am living near the dutch border, so I am at least halfway competent ;-)
350	manuzhai so what are we actually talking about? :P
351	rmunn Only Guido van Rossum (and his brother) are smart enough to understand some of this stuff. :-)
352	adrian_h hugo-: What do you think of the helper function idea above?
353	hugo- adrian_h: if you find a better name ;-)
354	adrian_h create_model_from_dynamic_field_list_for_hugo()
355	adrian_h :)
356	hugo- lol
357	rmunn That's what, 30 characters long?
358	rmunn 40.
359	adrian_h At least this isn't PHP, where there's actually a performance penalty for using long variable names
360	Boffbowsh so glad to be moving over to python from php
361	Boffbowsh had to do a bit of php today, felt so... spaghetti like
362	adrian_h manuzhai: We're debating this syntax: http://django.pastebin.com/339441
363	Tybstar why the hell does viewcvs want TK??
364	* rmunn boggles. TK?
365	adrian_h The current question on the table is whether "class Meta" is a good name. It would conflict with a field named "Meta"
366	hugo- hmm. maybe if the mechanisms are made more transparent - define a set of functions to construct a model and then define the transformations that are used for transforming the class definition using _those_ functions into a model
367	manuzhai adrian_h: how is your current position?
368	adrian_h manuzhai: I'm getting more and more interested in the new syntax
369	hugo- that way the new syntax works nicely and somehugo with the need for programmatically creating models can use the documented "lower level"
370	adrian_h hugo-: That's a good solution
371	adrian_h A lower-level API for creating models
372	manuzhai adrian_h: that's good
373	hugo- actually that's the way Lispers often do this: build a set of base functions and define new syntax on top of it. They have macros, we have metaclasses for the syntax stuff
374	rmunn mmarshall's patch would need some serious rewriting, then.
375	Tybstar this is the suck.
376	rmunn Tybstar: Ouch. What distro?
377	Tybstar rmunn: it's a fedora machine
378	Tybstar i wish viewsvn worked
379	Netminder there's your problem :)
380	Tybstar heh
381	hugo- yikes. that's even worse than debians often-weird-dependencies :-)
382	* Tybstar wishes slashzero were here
383	Tybstar i just can't quite get viewsvn working
384	BleSS why i get this error? AttributeError: 'NoneType' object has no attribute 'status_code'
385	* rmunn hearts Synaptic
386	adrian_h BleSS: Your view has to return an HttpResponse object
387	* Notify: jacobkm is online (kornbluth.freenode.net).
388	* jacobkm (n=jacob@74.57.cm.sunflower.com) has joined #django
389	Boffbowsh BleSS: your get_object or whatever returned no object
390	Boffbowsh hence the Nojne
391	Boffbowsh *None
392	adrian_h That's a good case for a better error message.
393	Boffbowsh ahh, ok
394	rmunn adrian_h: Special-case check to see if None was returned from the view?
395	adrian_h rmunn: Yes
396	mmarshall rmunn: My patch is very simple; even stupidly simple now that the no-name-for-ForeignKeys thing is no longer needed.
397	hugo- since None is the implicit value for functions that are missing the return alltogether, I think that special case check would be usefull
398	Boffbowsh nn all
399	* proteusguy (n=proteusg@dsl027-163-201.atl1.dsl.speakeasy.net) has joined #django
400	* Boffbowsh has quit ()
401	* Dagur (n=dagur@dsl-196-5.hive.is) has left #django
402	* cmlenz has quit ()
403	BleSS adrian_h, whatever view has th reuturn an HttpResponse object? (my view simply add data to data base)
404	manuzhai BleSS: you should redirect then
405	adrian_h BleSS: Yes, every view has to return an HttpResponse object, so your users actually get a page. If you want to redirect, use HttpResponseRedirect('/path/to/new/page')
406	hugo- ok, so if the new syntax will be a combination of a documented model-creation API plus a compact class-based syntax on top of it, it get's a +1 from me
407	BleSS thanks
408	adrian_h hugo-: Yay
409	hugo- that should satisfy both the hacker and the maintainer in me ;-)
410	* jvoorhis (n=user@cpe-24-93-226-82.neo.res.rr.com) has joined #django
411	jvoorhis adrian_h: hi
412	adrian_h Hi jvoorhis
413	jvoorhis good work on the tutorial
414	* ljoramo (n=ljoramo@69-25-223-154.acsol.net) has joined #django
415	adrian_h jvoorhis: Thanks -- is everything going smoothly for you?
416	jvoorhis yes, except for python gotchas
417	adrian_h ah
418	jvoorhis a tuple is like a constant array, i guess
419	Netminder the sooner we make the change the better IMHO
420	adrian_h Netminder: Yes
421	adrian_h jacobkm: Hello
422	manuzhai adrian_h: could you or jacobkm look at my patch for #289?
423	jacobkm Howdy, adrian
424	jacobkm Looks like the logging bot has been broken for a few days :(
425	adrian_h Yeah :-(
426	rmunn Who set up the loglibrary logging in the first place?
427	jacobkm I did
428	rmunn So you're the one who can fix it (if it can be fixed).
429	jacobkm Yup
430	jacobkm I'm trying to figure out what's wrong right now.
431	rmunn I've got to go soon, but I'll save a copy of this IRC discussion in case anyone wants to look at what was said about #122.
432	Netminder I think loglibrary is having issues
433	Netminder I don't think there's anything we can do
434	jacobkm Hunh
435	rmunn Actually, most of it has already scrolled off the top of my scrollback buffer.
436	jvoorhis adrian_h: one thing i did not like was that the form on the details template has to be aware that it lives at polls/
437	jacobkm I was afraid of that.
438	rmunn Did anyone else manage to save the discussion from earlier?
439	manuzhai does anyone else think the CherryPy page on the wiki is pointless?
440	adrian_h manuzhai: Yeah, I just saw that and am tempted to delete it
441	hugo- rmunn: I can put the log from the discussion as a comment on #122 if you want
442	manuzhai I am, too
443	adrian_h rmunn: I've got the discussion
444	manuzhai please do so
445	manuzhai it seems like FUD
446	hugo- my scrollback buffer is big enough
447	rmunn Yes, please do.
448	jacobkm CherryPy page: gone
449	manuzhai hugo-: an attachment to #122 might be better
450	mrproper What is a closure & how does one use a closure in Python? (I have an idea of closure from Javascript and event listeners)

Download in other formats:

Original Format

Code

Ticket #122: modeldiscuss.txt

Download in other formats: