On Fri, Jul 17, 2009 at 11:57 AM, Bill Hart<[email protected]> wrote:
>
> As will have been observed, I've been thinking hard over the last few
> weeks as to how we can make MPIR contribution easier. In short it
> would be good to have more core developers. In particular I've been
> working on, or planning:
>
> * Set up a Git repo and encourage its use
> * Write developer documentation
> * Write an FAQ
> * Write an MPIR digest detailing development progress every three weeks
> * Write an MPIR most wanted list, of most needed contributions
> * Write a set of scripts or a program which makes adding new files to MPIR 
> easy
>
> In this post I'll concentrate on the last of these which I've been
> looking into for a couple of days.
>
> The idea I had was to begin work on a cross platform windows program
> for MS Windows, OSX, KDE, etc, which would take the pain out of adding
> new files and modules to existing C maths libraries. With a little
> configuration it would essentially write all the boilerplate for the
> user, including stubs for test code, the function being added, and
> basically do all the configure and make stuff that needs to be done
> automatically.
>
> I decided to look into writing a basic programmer's editor, linked
> with Git, which would have this extra functionality of adding
> boilerplate code for new functions. The first step was to write a
> basic windows editor with C syntax highlighting and block folding.
>
> As python is cross platform I looked for libraries for Windows
> development. Initially I found:
>

Did you look at PyQt? I haven't used it for a long time, but it used
Qt's Designer which worked pretty well and I think PyQt includes the
Scintilla text editing library which does syntax highlighting, etc.

> * Python 2.6, which is cross platform
> * pygments - for syntax highlighting text, outputting it to rich text format
> * wxPython - a python port of wxWidgets, a cross platform windows library
> * BoaConstructor - a RAD tool for fast development of wxPython code,
> with a GUI designer and IDE
>
> Here are my notes per package:
>
> * Python 2.6 - installs fine on my Windows box. Has a command line and
> a windows editor. No problems with this as far as I know.
>
> * Pygments - is distributed as a python egg. There is a version for
> python 2.6. So I look up how to install a python egg. Apparently the
> easiest way is to use easy_tools. To get that you have to get
> setup_tools. There is no Windows version of this for python 2.6. But
> you can get setup_tools for python 2.6. It is distributed as a python
> egg. Arggh!!
>
> So I uninstal python 2.6 and install python 2.5. I install setup_tools
> and pygments. If I need to use it, I probably can, but see below, as
> stc may be a better option.
>
> * wxPython - installs fine and appears to work. Implements a styled
> text control (stc) widget which has a built in lexer for C++ and the
> ability to syntax highlight C++. It also provides options for block
> folding. It has essentially been designed specifically for writing a
> programmer's editor with all the standard features. Documentation
> however, sucks. In order to implement syntax highlighting, one needs
> to make use of options such as STC_C_COMMENTLINE, which as far as I
> can find, are basically undocumented.
>
> In fact, there is basically no documentation on the wxPython website
> for the stc control. However someone has gone to the trouble of
> attempting to document it here:
>
> http://www.yellowbrain.com/stc/index.html
>
> However, the meaning of the various stc variables is not listed. Only
> a list of possible variables is given, here:
>
> http://www.yellowbrain.com/stc/varwrap.html
>
> The stc also provides a lexer for asm syntax highlighting, though I
> have no idea which asm format it highlights. Again that is
> undocumented, and a google search does not help with finding proper
> documentation for any of the stc.
>
> * BoaConstructor - Installs fine. But full of bugs.
>
> 1) You have to put all graphical widgets down in precisely the correct
> order first go in the GUI designer, otherwise you have to do your
> entire project from scratch, as there is no easy way to change it once
> it is down. For example you can't remove a panel and replace it with a
> sash window if at some point you change your mind.
>
> 2) When any widgets are moved around, they do not render properly. You
> can't see them or they cause blag drag marks across all the other
> widgets.
>
> 3) There's no easy way to select a widget to delete it. The only way I
> have found is to click on it in the inspector, hope that some black
> sizer dots appear, click precisely on one of those dots, then click
> delete. There's no way to graphically select the widget you want to
> delete.
>
> 4) It's impossible to move some widgets. Even selecting them via the
> workaround in 3 does not allow one to move the widgets, as you cannot
> move them by clicking on the sizer dots, you have to click in the
> widget's center, which causes another random widget to be selected.
>
> 5) BoaConstructor crashes frequently, losing your work. And I don't
> mean that python just gives some kind of error message. The whole
> thing actually segfaults and drops you back to the desktop.
>
> 6) If you rename any of the default names, like frame1, that
> BoaConstructor gives the widgets, it loses track of them and you have
> to start your entire project from scratch. This is irrespective of
> whether you rename them in code, the inspector or otherwise.
>
> 7) It is very difficult, though not impossible to figure out how to
> associate menus with options on menubars. The logical way of doing
> this via the inspector has not been implemented.
>
> 8) Many widgets come up with default sizes of zero, or panes of size
> zero. Thus there is no way to see them, resize them, add things to
> them, etc.
>
> 9) BoaConstructor regularly loses the connections between various
> widgets, e.g. if you add a sash window into a split window and a text
> control the other side of the sash in the split window, the inspector
> frequently loses the association.
>
> 10) BoaConstructor checks your code for you, and helpfully prevents
> you from entering the GUI designer if you have written correct code,
> by telling you that you have supplied the incorrect number of operands
> to various functions. If you supply the incorrect number as it wants
> you to, it either doesn't compile, or BoaConstructor simply thinks you
> still have the incorrect number of operands.
>
> So quite clearly BoaConstructor is still far too underdeveloped to be
> stable. It's only a 10 year old project. Using it is pointless.
>
> So perhaps I made the wrong choice. What else is available for GUI
> design which operates with xwPython?
>
> Two other highly recommended options are wxGlade and PythonCard. I
> tried the latter.
>
> * PythonCard - The Windows installer failed. I went back to an earlier
> version of python card. The Windows installer failed. This tells me
> they have precisely zero users on Windows, and they don't know about
> it. Eventually I got the source. I opened the documentation and went
> to the install instructions. I clicked on Windows install and got a
> dead link. I clicked on OSX install and got a dead link. I eventually
> found a document somewhere telling me how to install on some weird OSX
> variant. I got just enough info from that to tell me how to install
> PythonCard.
>
> I started reading how to use it. Instead of being an IDE, it is a DE,
> i.e. it is not integrated at all, but is a serious of completely
> unrelated tools. The source code it emitted was also not doing import
> wx, as I expected, but import card. So it is implemented as a library
> on top of wxPython. Given that this was next to useless I gave up.
>
> * wxGlade - admits on their website that it is not an IDE, but simply
> a designer and the generated code only displays the widgets, and no
> more. It recommends that people after an IDE use PythonCard,
> BoaConstructor or spe.
>
> So I look into that last option:
>
> * spe - the website is shocking. I couldn't make head nor tail of it.
> It appears that you have to download the source code from
> subversion.... Haven't tried that out yet.
>
> So I backtracked at this point and thought, perhaps wxPython is not
> the best choice for a widget toolkit for cross platform windows
> development.
>
> Everywhere I look however, I see two main options recommended.
> wxPython and TkInter, the standard GUI toolkit distributed with
> Python. Apparently every year it is a standing tradition to affirm
> TkIinter as the standard GUI library distributed with Python, however
> numerous people think it is dead, because of wxPython.
>
> Well maybe python is the wrong language for this sort of thing. But
> what else is there Java? Yuck. C++, too much work, though wxWidgets is
> available for C++ and was probably available for C++ first. But I've
> looked for decent RAD and GUI designers for C++ before, and all the
> good ones are commercial.
>
> Probably these days, the hot area to develop such cool tools is over
> the wire, i.e. browser based stuff. But I don't want to reinvent the
> wheel, and I have little experience with anything web related other
> than Javascript, which is far too slow and difficult to code, from
> experience. Too many systems to learn there otherwise.
>
> This is proving to be too frustrating. I'm going to look into spe, and
> also see what the canonical option for RAD/GUI design with TkInter is,
> including looking for an already implemented widget for source code
> highlighting. If I don't find what I am looking for, I'm officially
> giving up on the windows route.
>
> That would leave bash scripts maybe, or a command line C or python
> program. But I've no idea how to make the very complex feature I want
> to eventually implement, work from a command line interface. If I
> start this project, I obviously want it to go a lot further than just
> allowing one to add a few files to maths libraries. I want to add
> parsers which will allow for simplification of various repetitive and
> boring things we have to do over and over again when writing C maths
> libraries. At the very least it will need to parse C code and assembly
> code, but there's a whole lot more to what I had been thinking
> through.
>
> Bill.
>
> >
>

--~--~---------~--~----~------------~-------~--~----~
You received this message because you are subscribed to the Google Groups 
"mpir-devel" group.
To post to this group, send email to [email protected]
To unsubscribe from this group, send email to 
[email protected]
For more options, visit this group at 
http://groups.google.com/group/mpir-devel?hl=en
-~----------~----~----~----~------~----~------~--~---

Reply via email to