Re: [GSoC][Patch] Automatic Mode Detection V1

From: Yosry Muhammad <yosrym93(at)gmail(dot)com>
To: Dave Page <dpage(at)pgadmin(dot)org>, pgadmin-hackers(at)postgresql(dot)org
Subject: Re: [GSoC][Patch] Automatic Mode Detection V1
Date: 2019-06-19 13:47:04
Message-ID: CAFSMqn8X6FgRM98xXm_cyyWonQ5j-n_O_HCkL4j3Daw=4ATuEw@mail.gmail.com
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgadmin-hackers

Hi,

On Wed, Jun 19, 2019, 1:54 PM Dave Page <dpage(at)pgadmin(dot)org> wrote:
.

- We need to add a "do you want to continue" warning before actions like
Execute or EXPLAIN are run, if there are unsaved changes in the grid.

- I think we should make the text in any cells that has been edited bold
until saved, so the user can see where changes have been made (as they can
with deleted rows).

Both done, new rows are highlighted too.

Nice! I realise it's most likely not your code, but if you can fix the
wrapping so it doesn't break mid-word, that would be good. See the attached
screenshot to see what I mean.

Will do.

- If I make two data edits and then delete a row, I get 3 entries in the
History panel, all showing the same delete. I would actually argue that
data edit queries that pgAdmin generates should not go into the History at
all, but maybe they should be added albeit with a flag to say they're
internal queries and an option to hide them. Thoughts?

That was a bug with the existing 'save changes' action of 'View Data', on
which mine is based upon. I fixed the bug, now the queries are shown
correctly. However, the queries are shown in the form in which they are
sent from the backend to the database driver (without parameters - also an
already existing bug in View Data Mode), for example:

INSERT INTO public.kweek (
media_url, username, text, created_at) VALUES (
%(media_url)s::character varying, %(username)s::character varying,
%(text)s::text, %(created_at)s::timestamp without time zone)
returning id;

I propose two solutions:
1- Hide pgadmin's generated sql from history (in both modes).
2- Show the actual sql query that was executed after the parameters are
plugged in (more understandable and potentially helpful).

I like the idea of doing 2 - but I think we should have a checkbox on the
history panel to show/hide generated queries (and we should include
transaction control - BEGIN, COMMIT etc - in the generated query class).

I can work on option 2 now and then work on
the checkbox if/when there is time.

- We need to think about how data editing fits in with transaction control.
Right now, it seems to happen entirely outside of it - for example, I tend
to work with auto commit turned off, so my connection sits
idle-in-transaction following an initial select, and remains un-affected by
edits. Please think about this and suggest options for us to discuss.

I integrated the data editing in the transaction control as you noted. Now
the behavior is as follows:
1- In View Data mode, same existing behavior.
2- In Query Tool mode:
- If auto-commit is on: the modifications are made and commited once save
is pressed.
- If auto-commit is off: the modifications are made as part of the ongoing
transaction (or a new one if no transaction is ongoing), they are not
commited unless the user executes a commit command (or rollback).

That seems to work. I think we need to make it more obvious that there's a
transaction in progress - especially as that can be the case after the user
hits the Save button and thinks their data is safe (a side-thought is that
perhaps we shouldn't require the Save button to be pressed when auto-commit
is turned off, as that's just odd). We should highlight the transaction
state more clearly to the user, and make sure we prompt for confirmation if
they try to close the tab or the whole window.

The transaction status can be made more obvious and point out when a
transaction is in progress that changes aren't commited. However, removing
the save button when auto commit is off will cause us to a send a request
and execute a query every time any cell is changed (which can be by
accident or some kind of draft). I also think it will make more sense when
there is a dedicated button, which can be named such that it is clear that
it only executes some queries. Also, the pop up that shows after edits are
succeesful can also state thar these changes are not yet commited.

I think it makes more sense for filters to be disabled. I mean since the
user is already writing SQL it would be more convenient to just edit it
directly.

Well we're not going to just disable them - we'll either remove them, or
try to make them work. I'm leaning strongly towards just removing that code
entirely.

I meant disabling them in the query tool while keeping them in the View
Data mode as the user cannot edit the sql in the View Data mode. Do you
want to remove the feature from both modes completely?

I think you misunderstand - I want to remove the View Data mode entirely.
Your work should replace it.

As a user of pgAdmin I think this might not be the best option. Not all
users of pgAdmin are developers or know SQL. I worked on several projects
before where other people on the team (or frontend developers) would just
want to take a look at some data or do simple edits using the GUI. Also,
other management studios for other DBMSs also allow for this. In addition,
the user can do sorting of data without knowing SQL. What I think can be
done (potentially - maybe in the future) is limit the dependance on SQL
knowledge when doing filters in View Data mode, while disabling filters and
so in the Query Tool.

Thanks !

In response to

Responses

Browse pgadmin-hackers by date

  From Date Subject
Next Message Dave Page 2019-06-19 14:10:51 Re: [GSoC][Patch] Automatic Mode Detection V1
Previous Message Dave Page 2019-06-19 11:54:43 Re: [GSoC][Patch] Automatic Mode Detection V1