On Friday, 4 May 2018 at 19:54:46 UTC, Matthias Klumpp wrote:
I've written an email to Vadim, maybe we get a reply on the
status of both projects.
On Friday, 4 May 2018 at 07:18:09 UTC, bauss wrote:
[...]
Would it maybe be easier for you to base on ddbc[1] or
another existing abstraction layer for database abstraction?
Ddbc is pretty neat, and even has support for reading structs
directly from the database.
Perhaps, but it'd have to be a forked version as I don't
really want to depend on something that isn't updated
regularly.
ddbc's last commit was a year ago.
Yes, at the moment using ddbc and relying on it would mean
taking over some maintenance of it. Ddbc is fairly complete
though and might save you a lot of work, because it already
abstracts a lot of (relational) database systems.
I can't seem to find a license for it though?
It's Boost licensed (BSL-1.0), but probably needs an explicit
LICENSE file.
Maybe we can move ddbc to dlang-community, so more people can
easily commit changes to it (provided it gets accepted there,
and Vadim agrees with that move as well).
Perhaps I will end up having another "optional" dependency
to it as a temporary until I can have a better
implementation or something.
Taking over maintenance of it might be easier than
reimplementing the database abstraction again though.
For Postgres, ddbc worked really well for me, and I assume its
SQLite, MySQL and ODBC drivers are also still working well,
meaning less work for you.
Without ddbc, you'd have to write new abstraction on top of
some other libraries, like dpq2.
The frontend part of postgresql is almost finished, it's
just having the postgresql driver working properly, which is
where it's frozen right now.
Hmm... Does any public code for that exist already that I
could play around with?
Unfortunately, I have a few more unusual requirements for
Postgres, like:
* UUIDs as primary keys, instead of integers
As far as I remember the implementation of @DbId in Diamond,
then it supports whatever type.
Native support for std.uuid.UUID would be neat :-) For
Hibernated I use a mixin to convert a UUID into strings
transparently, for database insertion.
Diamond doesn't care much about what type your primary key is.
I will make sure that's how it function of course, if it
currently doesn't behave like it, but I'm pretty sure it does.
* Ability to register custom datatypes with the ORM (version
numbers in this case, the ORM can view them as text, but the
database has a special type for them)
That could be done with some attribute that lets you handle
columns yourself.
Do you have a good name for it?
I was thinking @DbProxy and then the function would be
something like:
[...]
Registering a new type with Postgres yields a new OID to
identify the type, so I would need a function to tell the
Postgres backend to treat OIDs of a certain number like "text"
types. Ideally, I would also need to annotate entity to set a
specific column type, like:
```
class Entity {
UUID uuid;
@ColumType("versionnumber")
string _version;
}
```
(With the result of CREATE TABLE not setting a "text" type for
the new version column, but a "versionnumber" type instead)
That would do it. For Hibernated, I have Hibernated create the
table initially, and then fire an ALTER TABLE at it afterwards
to change the column type, while at the same time registering
the new type OID with ddbc to be treated as text.
As said, this is a very specific requirement very few people
will have ^^
Much more frequently people will ask for JSONB and JSON type
support for a postgres driver though, I guess. For that,
specifying the column type explicitly could be quite helpful as
well, so switch between JSONB and JSON.
* Obviously the usual ORM stuff, one-to-many, many-to-many,
etc. relations
Yes, relations is one thing I haven't added and I have been
wanting to do it for a while.
I will definitely look into having it added as well.
That's kind of key for an ORM :-) Handling relations manually
was what made me abandon my "I just write raw SQL for
everything" ways, because it gets quite complex and annoying in
the long run.
(Obviously not a must-have list, I added support for custom
datatypes to my ddbc fork as well, because it's not really a
feature many people need)
Well, that's kind of the key to most of the development in
Diamond.
I usually add functionality that isn't widely used and in most
cases people implement it themselves ex. the whole diamond.seo
is not usually something a framework has.
I keep an eye on it - at the moment, Vibe.d satisfies all
requirements I have on a web framework, but that might change.
A well integrated ORM would certainly be a game changer, since
Vibe.d is limited to Mongo and Redis only.
Diamond is a neat project, I played around with it about half
a year ago, but didn't test the ORM part at all back then.
It wasn't that good back then and has improved a lot since, as
well many other parts of Diamond. Over the past half year it
has grown rapidly, both in stability and functionality.
It's quite an impressive piece of work, especially since it
looks like you're the only one working on it.
Keep it up! :-)
Thanks for the license information. I wouldn't mind taking over
maintenance of ddbc and you're right that it would definitely be
much easier to integrate. Also it could be better to cut down
different dependencies for each driver.
Let's see what we can figure out. I have a lot of this weekend so
I'll try to throw in time for this.
And yes, I've been working as the only developer since I started
two years ago, but sometimes it would be nice with more
developers as development would be faster and it wouldn't die
when I'm busy
