Each session has internal cache (a.k.a IdentityMap) of all loaded objects, and each object is loaded exactly once. New session has new cache, independent from the previous one, because the cache of a previous session is already obsolete

Yeah so why not make that IdentityMap global, i.e. not inside the session, and just reload the element if it is still around?

I mean the IdentityMap could be a WeakRef, so if an entity it isn't used anywhere else, *poof*, It's gone

I mean the IdentityMap could be a WeakRef, so if an entity it isn't used anywhere else, *poof*, It's gone

Maybe for the current session that should not happen, but for old sessions, if the object isn't stored in any other variable, it won't be used ever again

Therefore it could get garbage collected

You could call it “foundit”, like when you loose your car keys ?

Lol

We need to invalidate objects, because they may be updated inside the database by another concurrent processes. When working with database we use transactions. Each transaction has independent "world view" and work with the database state as if nobody else change the database at the same time. After transaction is finished, we cannot assume that the objects are in the same state, because they may be also updated by concurrent transactions. There is no easy way to listen database and invalidate all concurrently updated objects in memory. We can have a global long-lived cache which eliminates the need of db_session, that is basically what interactive mode does. But in this case you will encounter one day that you are working with obsolete data. Also, without db_session your global cache will load indefinitely and will not be garbage collected db_session implements standard architecture patterns Unit Of Work https://martinfowler.com/eaaCatalog/unitOfWork.html and Indentity Map https://martinfowler.com/eaaCatalog/identityMap.html and provide a convenient way of implicit working with session cache without the necessity to do session_cache.attach(obj)

Unfortunately IndentityMap cannot be a weakref, because objects in IdentityMap are interlinked. Consider the following example: with db_session: s1 = Student[123] for s in s.group.students: print(s.name) Here group.students holds references to all students of the group, and each student.group refers to g object As we hold a reference to one student, we indirectly hold all other objects too, via circular references. If we use weakref for circular references, the group and all students beside s1 will be garbage collected immediately. So we need to use regular references in IdentityMap instead of weakrefs

> but for old sessions, if the object isn't stored in any other variable, it won't be used ever again It is not correct, as object can be accessible via relationships: s1.group.students

hmmm

It would help replying to you if you'd hit enter after paragraphs so I can reply to them individually.

Ok

maybe you can just copy paste the text, if you want to write it first

> As we hold a reference to one student, we indirectly hold all other objects too, via circular references. If we use weakref for circular references, the group and all students beside s1 will be garbage collected immediately. So we need to use regular references in IdentityMap instead of weakrefs Can we make just the reverse elements weak? Like | Student | Group | ---------------------- | Günter |  12 | | Herbert |  12 | | Amadeus | 43 | So we could have Student.group normal, but Group.students weak?

That would be the way we would need to load them from the database anyway, I imagine

If you load Group, it wouldn't load all .students

but with a Student, the Group would be non-weak, and don't get deleted.

Also, for the current session, all entities would be stored non-weak Indentity Map, only when an new session is started they get transfaired into a weak global version

> Can we make just the reverse elements weak? > So we could have Student.group normal, but Group.students weak? If a user does: g = Group[123] for s in g.students: print(s.name) he want to be able to traverse relationship from group to student, and to do this group.students should contain a set of students, which will not be garbage-collected immediately

If you load Group, it wouldn't load all .students

But if you access group.students, you do

Hmmm. Damnit

Also, for the current session, all entities would be stored non-weak Indentity Map, only when an new session is started they get transfaired into a weak global version

Then you cannot render group with group.students in Jinja template after db_session is over

How are those garbage collected anyway?

After db_session is over, it disconnect objects from itself, but they remain interlinked with each other via references. If there is a reference to at least one object, it hold all interlnked objects. After all references from the outside to this group of objects are removed, Python garbage-collect them. Python can detect cross-referenced objects which are not referenced from the outside

For db_session(strict=True) all relationships between objects are removed, so Python can garbage-collect objects more efficiently

I'm still not entirerly sure we can't use that process and just shove them into an additonal weak list to recycle the existing objects?

The main point of db_session is to denote the moment when the objects can be garbage-collected when no outside references to them exists

I'm still not entirerly sure we can't use that process and just shove them into an additonal weak list to recycle the existing objects?

I'll think about it, but it does not look easy to implement and robust

Especially in multi-threaded applications

No it doesn't

that is true

Hello https://editor.ponyorm.com/user/<username> does not fully load for me, does it work for you?

Failed to load resource: net::ERR_CONNECTION_REFUSED jquery-3.1.0.min.js:1

It works for me, maybe some internet connectivity problem

everything else on the internet works, but thanks

Hi! How install Pony 0.7.7?

Pip install git+https://url@orm#pony

Url is git clone url

Hi guys and also those rare girls! @metaprogrammer not sure who's responsible for that, but I just switched to Basic plan for only $5 per this month. I mean an editor.

Hi Anzor! Do I understand correctly that you've switched from Free to Basic plan on editor.ponyorm.com and everything is ok?

Hi Anzor! Do I understand correctly that you've switched from Free to Basic plan on editor.ponyorm.com and everything is ok?

Yep

https://github.com/stsouko/LazyPony

https://github.com/stsouko/LazyPony

Permanent link to the stsouko/LazyPony project you mentioned. (?)

Wow nice picture

this is old avatar. I don't know why it not changed

Hey. Can you give links to big projects where pony is used? I want to study the code.

https://github.com/Tribler/tribler

https://github.com/Tribler/tribler

Permanent link to the Tribler/tribler project you mentioned. (?)

thank!! I will go to study)

Hello! how to set old style sql_debug for to see all queries in every threads/functions/methods?

set_sql_debug(True)

old sql_debug(True) works too, but will be deprecated

Oh, you mean, in all threads

I wrote mixins for Entity. in this mixins I do many queries. but I can see only top level queries

I think generators ignore sql_debug

I'll check that today. Right now you can do import pony.orm.core and then pony.orm.core.local.debug = True before queries as a quick fix

Thank You!

not working to me

or before all queries?

To be sure, you can put it before each query

Are you sure queries are actually executed?

To be sure, you can put it before each query

this also not working. Yes. I see results of execution. in debug mode line by line also

Maybe you use logging module and it is misconfigured, so it suppress output

pure python main.py also not working. only first query showed

That looks strange

Do you have import logging in main.py or in some module which imports from main.py?

only this from logging import warning

can you remove it?

Actually, it doesn't matter

Actually, it doesn't matter

yep

> in debug mode line by line also if you use debugger you can put bereakpoint inside pony.orm.core.log_sql function (it's on core.py:112) and see if it executed

yes. it executed

for every

but not printed

if has_handlers(sql_logger): this condition executed

level 20 is disabled

Ok, so the logging module is configured to silent INFO-level messages

You need to change it

Or you can direct pony to use another level for its messages

import pony.orm.core pony.orm.core.orm_log_level = logging.WARNING But probably it is better to set logging level to INFO