Re: PATCH: Debugger Redesign

From: Ashesh Vashi <ashesh(dot)vashi(at)enterprisedb(dot)com>
To: Dave Page <dpage(at)pgadmin(dot)org>
Cc: pgadmin-hackers <pgadmin-hackers(at)postgresql(dot)org>
Subject: Re: PATCH: Debugger Redesign
Date: 2013-04-11 17:10:23
Message-ID: CAG7mmoyj14zgVk3R6+SyJazfdn0x2ehX9F7pAc8agmDmR=nF-w@mail.gmail.com
Views: Whole Thread | Raw Message | Download mbox | Resend email
Thread:
Lists: pgadmin-hackers

Sure.
I'll work on it...
On 11 Apr 2013 21:00, "Dave Page" <dpage(at)pgadmin(dot)org> wrote:

> Hi
>
> First impressions - it already seems more stable than the old code,
> and I only played with it for an hour or so on Mac so far. Nice work
> :-)
>
> I did find a few things in my initial testing that need some
> attention. I didn't get the impression anything was particularly
> serious or would be troublesome to fix though:
>
> - The layout of the parameters dialogue needs work - the grid needs to
> size to the dialogue.
>
> - The checkboxes in the grid are jumbo sized. We have them in the Edit
> Grid already - can the code be reused?
>
> - The popup activity dialog is kinda annoying. Perhaps we could put a
> progress indicator in the status bar? Low priority.
>
> - When injecting new values into variables, if the value can't be
> accepted (wrong datatype etc), then the grid should be reset to the
> original value when the error message box is accepted.
>
> - If a function is re-executed in the same session, the return value
> isn't cleared from the Results grid.
>
> - If a function is re-executed in the same session, the breakpoints
> are cleared. This doesn't seem to happen with in-process debugging,
> only direct.
>
> - Aborting a direct debug session mid-function caused an indefinite(?)
> hang with a busy cursor.
>
> - The status messages in the status bar are a little confusing. I
> think you need to add "Done." to the end of the string when an action
> completes, otherwise it looks like things never complete until you
> step or run.
>
> If you can fix those issues (with the possible exception of the
> progress indicator), I'll dive in a bit deeper. I don't want to go too
> deep just yet in case you have to change more than I expect.
>
> Thanks.
>
> On Wed, Apr 10, 2013 at 12:10 PM, Ashesh Vashi
> <ashesh(dot)vashi(at)enterprisedb(dot)com> wrote:
> > Hi All,
> >
> > We always wanted to redesign the plpgsql debugger in pgAdmin III from
> > very long time due to very long list of bug
> > Reports of the debugger. We were not able crack down variety of cases
> across
> > the platforms. For many cases - resolving
> > On one platform, we ended up introducing other issues on another
> platforms.
> > This inability of handling variety of cases
> > Across platforms made us to redesign the debugger from the scratch.
> >
> > Let me list down few the limitations/flaws with the current
> implementation
> > of the debugger:
> > - Uses its own mechanism for making connection, querying (async/sync) &
> > handing results instead of using the common
> > available classes, which are used by all other components in pgAdmin
> III.
> > i.e.
> > dbgConn, dbgResult, dbgPgThread, etc.
> >
> > Because of this, the improvements in the existing infrastructure were
> > never reflected in the debugger.
> > i.e.
> > The connection with debugger were never be created with the SSL
> support.
> > Because - dbgConn never supported the SSL.
> > This is also results for the many bugs.
> > i.e.
> > The dbgPgThread has its own mechanism to run queries asynchronously
> and
> > resulted into many bugs. It generates few
> > events, which shouldn't be done after the completion of the debugger
> > operations. And, that leads to crash. Using
> > TestDestroy() function for JOINABLE threads were unnecessary and never
> > required. But - it is using it.
> >
> > - A hidden windows is taking care of the events generated during the
> > debugging the actual window. It could have been
> > Good, if all the tasks having intialited and handled from by this
> central
> > mechanism, instead we do have different
> > places for handling both the direct/indirect debugging. And, that
> > introduced a lot of issues. And, that made the
> > things very confusing even to understand the existing mechanism.
> >
> > And, many more.
> >
> > We had to made changes in the existing infrastructure/classes to use it
> with
> > the debugger.
> > * pgConn:
> > + Report error quietly (if told) while error found in execution, this
> > allows these functions to be used from any
> > Thread. Otherwise, it used to throw an error.
> > + Set/Reset the PGcancel object before and after the query execution,
> > which will be useful for us to use the new
> > Mechanism of libpq (of course, not very new) to cancel any query
> > execution from in between.
> > + Allow new application name, while duplicating the connection
> > + Save the version and the features list, while duplicating the
> > connection, this will avoid/reduce the duplicate calls
> > To the database server.
> > * pgQueryThread:
> > + Allow to specify the custom notice processor and handler
> > + Allow to run multiple queries by creating queue
> > + Allow to use the EnterpriseDB's callable statement from the
> > pgQueryThread (if available)
> > + Proper cancellation mechanism implementation
> > + Send a custom event to the caller (instead of standard event) for
> > successful execution or on different errors
> > (but - not if it is explicitly cancelled by the user.) This allows
> us to
> > send more information to the developer.
> > + Allows the developer to add a batch queries (which should run
> > asynchronously) even after starting of the thread.
> > + Allow to use the PGparam now, to avoid generating the queries
> > dynamically
> >
> > The new design is based on the MVC (Model-View-Controller) architecture.
> > * Controller (dbgController)
> > - A central mechanism to handle all the operations.
> > * View (frmDebugger - mostly reused)
> > - Responsible for presentation and interaction with the end-user
> > i.e.
> > code viewer, different variable windows, stack window, etc.
> > - It also controls other part of view.
> > * Model (dbgModel)
> > - Handles/Contains all the data
> > - It contains the information about the target function/procedure.
> >
> > For any operation, View and Model will ask the Controller to handle it.
> >
> > Summary of the changes:
> >
> > INSERTED| DELETED| FILENAME |
> SUMMARY
> >
> ---------+----------+---------------------------------------------|----------------------------
> > 3| 1| pgadmin/ctl/ctlSQLResult.cpp | -
> Call
> > cancel thread execution on exit, if already
> > |
> running
> > for nice exit handling
> > 196| 18| pgadmin/db/pgConn.cpp | -
> > Set/Reset the PGcancel object
> > | -
> Saving
> > settings and new application name, while
> > |
> cloning
> > 31| 24| pgadmin/include/db/pgConn.h |
> > 245| 33| pgadmin/include/db/pgQueryThread.h |
> > 647| 111| pgadmin/db/pgQueryThread.cpp | -
> > Multiple queries execution support (one-by-one)
> > 8| 0| pgadmin/include/db/pgSet.h | -
> > Introduce a new function GetCommandStatus for the
> > |
> result
> > 0| 1| pgadmin/db/pgSet.cpp | -
> > Removed unnecessary header inclusion
> > 0| 1563| pgadmin/debugger/ctlCodeWindow.cpp | -
> > Omitted/Removed it completely
> > 0| 250| pgadmin/include/debugger/ctlCodeWindow.h |
> > 9| 9| pgadmin/debugger/ctlMessageWindow.cpp | -
> > Standardize the function naming conversion with the
> > | other
> > components
> > 5| 12| pgadmin/include/debugger/ctlMessageWindow.h |
> > 28| 27| pgadmin/debugger/ctlResultGrid.cpp | -
> Using
> > the standard querying mechanism (pgSet)
> > |
> instead
> > of libpq's PGresult directly
> > 4| 3| pgadmin/include/debugger/ctlResultGrid.h |
> > 4| 7| pgadmin/debugger/ctlStackWindow.cpp | -
> > Standardize the function naming conversion with the
> > | other
> > components
> > 2| 2| pgadmin/include/debugger/ctlStackWindow.h |
> > 50| 56| pgadmin/debugger/ctlTabWindow.cpp | -
> > Standardize the function naming conversion with the
> > | other
> > components
> > 20| 17| pgadmin/include/debugger/ctlTabWindow.h |
> > 60| 55| pgadmin/debugger/ctlVarWindow.cpp | -
> > Standardize the function naming conversion with the
> > | other
> > components
> > 17| 15| pgadmin/include/debugger/ctlVarWindow.h |
> > 819| 0| pgadmin/debugger/dbgController.cpp | -
> > Central of the debugger (it takes care of all the
> > |
> > operations, the frmDebugger (view) and dbgModel
> > |
> (model)
> > interact with it.
> > 173| 0| pgadmin/include/debugger/dbgController.h |
> > 0| 21| pgadmin/debugger/dbgDbResult.cpp | -
> > Omitted/Removed it completely
> > 703| 0| pgadmin/debugger/dbgEvents.cpp | -
> Added
> > new file, which contains all the event
> > |
> handling
> > functions of the dbgController to reduce
> > | the
> size
> > of dbgController and separating them from
> > | the
> > direct operations
> > 63| 0| pgadmin/debugger/dbgModel.cpp | -
> > Contains all the information regarding the
> > |
> target.
> > 143| 0| pgadmin/include/debugger/dbgModel.h |
> > 0| 544| pgadmin/debugger/dbgPgConn.cpp | -
> > Omitted/Removed it completely
> > 0| 363| pgadmin/debugger/dbgPgThread.cpp | -
> > Omitted/Removed it completely
> > 0| 157| pgadmin/debugger/dbgResultset.cpp | -
> > Omitted/Removed it completely
> > 536| 194| pgadmin/debugger/dbgTargetInfo.cpp | - It
> > fetches and saves the target information from
> > | the
> > database server (it is no more using the
> > |
> > deprecated pldbg_target_info function).
> > 146| 75| pgadmin/include/debugger/dbgTargetInfo.h |
> > 44| 101| pgadmin/debugger/debugger.cpp | -
> > Debugger menu factory, modified accordingly new
> > |
> design
> > 770| 682| pgadmin/debugger/dlgDirectDbg.cpp | -
> > Handles now only taking input from the end-user
> > | - It
> > also takes care of converting an expression
> > |
> input to
> > a proper value
> > | - It
> > also allows the end-user to use the default
> > |
> value of
> > an argument through UI.
> > | - It
> > does not control the debugger execution any
> > | more
> > 74| 48| pgadmin/include/debugger/dlgDirectDbg.h |
> > 734| 165| pgadmin/debugger/frmDebugger.cpp | -
> Mostly
> > reused
> > | -
> Works
> > as presentation layer and takes input from
> > | from
> the
> > end-user
> > | i.e.
> > set/reset break-points, changing parameter
> > |
> > values, showing the target code, showing
> > |
> > current parameter values, etc.
> > 114| 92| pgadmin/include/debugger/frmDebugger.h |
> > 3| 5| pgadmin/debugger/module.mk |
> > 5| 1| pgadmin/dlg/dlgClasses.cpp | -
> Call
> > cancel thread execution on exit, if already
> > |
> running
> > for nice exit handling (ExecutionDialog)
> > 4| 1| pgadmin/frm/frmEditGrid.cpp | -
> Call
> > cancel thread execution on abort
> > 2| 2| pgadmin/frm/frmQuery.cpp | -
> > Modified to use new custom event on query
> > |
> > execution completion
> > 2| 1| pgadmin/include/frm/frmQuery.h |
> > 1| 1| pgadmin/include/db/module.mk |
> > 79| 0| pgadmin/include/db/pgQueryResultEvent.h | -
> > Introduced a new custom event for handling the
> > | query
> > execution error/success (pgQueryThread)
> > 14| 21| pgadmin/include/debugger/dbgBreakPoint.h | -
> > Reformatted the break-point according to new design
> > 0| 38| pgadmin/include/debugger/dbgConnProp.h | -
> > Omitted/Removed it completely
> > 57| 10| pgadmin/include/debugger/dbgConst.h |
> > 0| 53| pgadmin/include/debugger/dbgDbResult.h | -
> > Omitted/Removed it completely
> > 0| 99| pgadmin/include/debugger/dbgPgConn.h | -
> > Omitted/Removed it completely
> > 0| 95| pgadmin/include/debugger/dbgPgThread.h | -
> > Omitted/Removed it completely
> > 0| 52| pgadmin/include/debugger/dbgResultset.h | -
> > Omitted/Removed it completely
> > 2| 6| pgadmin/include/debugger/module.mk |
> > 3| 6| pgadmin/include/precomp.h |
> > 7| 12| pgadmin/pgAdmin3.vcxproj | -
> Window
> > build script changes
> > 10| 28| pgadmin/pgAdmin3.vcxproj.filters |
> > 3| 3| pgadmin/ii/dlgDirectDbg.xrc | -
> Input
> > Arguments dialog change (changed the
> > |
> > dimensions.)
> > 44| 44| pgadmin/ii/xrcDialogs.cpp |
> >
> >
> > --
> >
> > Thanks & Regards,
> >
> > Ashesh Vashi
> > EnterpriseDB INDIA: Enterprise PostgreSQL Company
> >
> >
> > http://www.linkedin.com/in/asheshvashi
> >
> >
> >
> > --
> > Sent via pgadmin-hackers mailing list (pgadmin-hackers(at)postgresql(dot)org)
> > To make changes to your subscription:
> > http://stream.boxwatch.ru/mailpref/pgadmin-hackers
> >
>
>
>
> --
> Dave Page
> Blog: http://pgsnake.blogspot.com
> Twitter: @pgsnake
>
> EnterpriseDB UK: http://www.enterprisedb.com
> The Enterprise PostgreSQL Company
>

In response to

Responses

Browse pgadmin-hackers by date

  From Date Subject
Next Message Guillaume Lelarge 2013-04-14 08:43:57 Re: Next 1.16 point release?
Previous Message Dave Page 2013-04-11 15:30:36 Re: PATCH: Debugger Redesign