Bochs/bochs/README-wxWindows

261 lines
11 KiB
Plaintext
Raw Normal View History

Readme for wxWindows Interface
updated Thu Aug 29 10:43:37 EDT 2002
Contributors to wxWindows port:
Don Becker (Psyon)
Bryce Denney
Dave Poirier
Volker Ruppert
wxWindows Configuration Interface
The wxWindows port began in June 2001 around the time of Bochs 1.2.1. Dave
Poirier and Bryce Denney started adding a wxWindows configuration interface.
We made some progress, but stopped after a while. Then in March/April 2002
Bryce and Psyon revived the wxWindows branch and turned it into a (mostly)
usable interface. Psyon did most of the work to get text and graphics
working, and Bryce worked on event passing between threads, and keyboard
mapping.
The wxWindows window appears first. This WILL BE where you can choose a preset
configuration or create one from scratch, then edit the bochsrc options using
the GUI. Then, by clicking on Simulate:Start, you can start up the Bochs
simulation. The VGA display is implemented in wxWindows.
WARNING: Don't expect perfect code yet!
Regarding wxWindows versions, we started out with 2.2.9 but then found
a reason or two to switch to 2.3.2. (I have forgotten exactly what
they were.) In addition, I made a keyboard patch to wxWindows itself,
which they accepted for the next version, that allows us to get raw
keycodes out of the library in addition to wxwindows translated keys.
Bochs will work with or without it, but you'll find that the keyboard
mapping is much better with the patch, which is called
patches/patch.wx-raw-keycodes.
So the ideal wxWindows version is 2.3.2+patch.wx-raw-keycodes.
To apply the patch, go to the main wx____-2.3.2 directory and
type "patch -p1 < THIS_PATCH_FILE".
Bryce has been developing in linux, where a "configure --with-wx;make"
should work fine.
To build in MS VC++:
- edit .conf.win32-vcpp and add "--with-wx" to the configure line.
If you want different configure options from what you see, change them
too.
- in cygwin, do "sh .conf.win32-vcpp" to run configure
- unzip build/win32/wxworkspace.zip into the main directory.
For cygwin: unzip build/win32/wxworkspace.zip
or use winzip or whatever else.
- open up bochs.dsw, the workspace file
- edit project settings so that VC++ can find the wxWindows include
files and libraries on your system. I installed them in
d:/wx/wx232/include and d:/wx/wx232/lib. Specifically, edit
- Project>Settings>C/C++>Category=Preprocessor: include directories.
- Project>Settings>Link>Category=Input: additional library path.
- build
Note that the project is set up for wxWindows 2.3.2, and the only
configuration that I've used is called Win32 Debug. To use on
other wxwindows versions, you will have to change some of the names
of the libraries to include. Use the samples that came with that
version of wxwindows for reference.
What works right now (August 2002):
- cd to a directory with a bochsrc and disk images in it (I use dlxlinux)
- start up bochs. You get a small window with a File, Edit, Simulate,
Debug, Log, Help menus. The empty black rectangle is where the
simulation will be displayed.
- choose File:Read Configuration and select the bochsrc file you want
to read.
- use Edit:Floppy Disk * or Edit:Hard Disk * to tweak settings.
These are the only dialogs that have been written so far.
Eventually you should be able to edit all settings using dialogs.
- choose "Start" on the Simulate menu. Bochs starts running in the
You can pause it, resume it, or kill it on the Simulate menu.
Most toolbar buttons work. Mouse support does not exist yet,
and the configure button doesn't do anything and should be removed.
- You can pause, resume, and stop the simulation too. If you start
a new simulation, be aware that may not work quite right. The
initialization and re-initialization process needs work.
- That's all. The other menu options are not implemented except for quit.
To do:
- sketch the GUI pieces that need to be made, with text description of
basically what they should do. Later in this file you'll find lots
of examples of this.
- have a few people look over the sketches so that we don't waste lots of
time building dialogs that people will hate.
- code some dialog boxes in wxwindows. Just get it looking right at first, and
Bryce can help you to get it connected to the simulator using the event
infrastructure that he has made. Again, after one or two is done there will
be some good examples to look at.
- implement ASK_PARAM gui event for all types of parameters. Currently only
string params are supported, meaning that when you call ask_param() on
a string parameter, the GUI gives the user a chance to edit that parameter's
value. We will need to implement boolean, enum, numerical parameters, and
lists. Lists should probably be displayed as a table or a dialog box
with several things to select.
- debugger interface. There's lots of demand for a debugger interface
to Bochs, probably more demand than for a configuration interface. We
should talk about what this will look like, and when we have a plan we
can start sketching and coding dialogs for the debugger interface too.
- clean up the biggest memory leaks and init/cleanup code. The gui allows you
to kill the simulator and restart, but it doesn't do well after the first
time. Valgrind should help with memory leak debugging, though until
recently it couldn't run multithreaded programes.
- disk change dialogs for floppy and cdrom need work.
http://sourceforge.net/tracker/index.php?func=detail&aid=545414&group_id=12580&atid=112580
- with --with-wx and debugger, control-C in the terminal window kills the
process instead of just interrupting the simulation.
- mouse events not implemented
-------------
------------------------------------------------------
Random notes follow
Added some sketches. I'm thinking that the control
panel will be able to basically show one of these screens at a time. When
you first start you would see ChooseConfigScreen which chooses between the
configurations that you have loaded recently (which it would remember
by the pathname of their bochsrc). Whether you choose an existing
configuration to be loaded or a new one, when you click Ok you go to
the first configuration screen, ConfigDiskScreen.
Each of the configuration screens takes up the whole control panel window.
We could use tabs on the top and/or "<-Prev" and "Next->" buttons to make
it quick to navigate the configuration screens. Each screen should
probably have a Prev, Next, Revert to Saved, and Accept button.
The menu choices like Disk..., VGA..., etc. just switch directly to
that tab.
------------------------------------------------------
Notes:
events from gui to sim:
- [async] key pressed or released
- [async] mouse motion with button state
- [sync] query parameter
- [sync] change parameter
- [async] start, pause, stop, reset simulation. Can be implemented
as changing a parameter.
- [async] request notification when some param changes
events from sim to gui:
- [async] log message to be displayed (or not)
- [async] ask user how to proceed (like panic: action=ask)
- [async] param value changed
- make my thread sleep for X microseconds (call wxThread::sleep then return)
In a synchronous event, the event object will contain space for the entire
response. The sender allocates memory for the event and builds it. The
receiver fills in blanks in the event structure (or could overwrite parts)
and returns the same event pointer as a response. For async events, probably
the sender will allocate and the receiver will have to delete it.
implement the floppyA and floppyB change buttons using new event
structure. How should it work?
vga gui detects a click on floppyA bitmap
construct a BxEvent type BX_EVT_ASK_PARAM
post the event to the wxwindows gui thread (somehow) and block for response
when it arrives in the gui thread, show a modal dialog box
get the answer back to the simulator thread
right now, this is working ok within the simulator thread using
wxMutexGuiEnter/Leave. Still I'm going to change it so that the
siminterface.cc code builds an event structure and the gui code
fills in the blank in the structure, instead of the stupid
notify_get_int_arg stuff.
Starting and Killing Threads
When a detachable (default) thread finishes (returns from its Entry()
function), wxwindows frees the memory associated with that thread.
Unless the thread is never going to end, it is potentially dangerous to have a
pointer to it at all. Even if you try to "check if it's alive" first, you may
be dereferencing the pointer after it has already been deleted, leading to it
claiming to be alive when it's not, or a segfault. To solve this, the approach
used in the wxwindows threads example is to have code in the thread's OnExit()
method remove the thread's pointer from the list of usable threads. In
addition, any references or changes to the list of threads is controlled by a
critical section to ensure that it stays correct. This post finally
explained what I was seeing.
+-----------------------
| From: Pieter van der Meulen (pgmvdm@yahoo.com)
| Subject: Re: Thread Sample program - bug
| Newsgroups: comp.soft-sys.wxwindows
| Date: 2001-06-28 17:51:35 PST
|
|
| At 06:24 PM 6/28/2001, you wrote:
| >Hi,
| >I have wxWindows 2.2.7 (wxMSW) installed.
| >
| >I just found in the thread.cpp sample code this section:
| >
| ><code>
| >void MyFrame::OnQuit(wxCommandEvent& WXUNUSED(event) )
| >{
| > size_t count = wxGetApp().m_threads.Count();
| > for ( size_t i = 0; i < count; i++ )
| > {
| >===> wxGetApp().m_threads[0]->Delete(); <=====
| > }
| >
| > Close(TRUE);
| >}
| ></code>
| >The indecated line should probably rather have a
| >m_threads[i] rather than m_threads[0] .
|
| No, it should not, although it is not immediately obvious. When Delete() is
| called, the thread will eventually delete itself, but not before it calls
| MyThread::Exit(), which will remove itself from m_threads[] using
| wxArray::Remove(this). wxArray::Remove (RemoveAt) will compact the array to
| remove the element, it is now size-1. After this wxThread::Delete() returns.
|
|
| >I have have a further question to this:
| >Does this mean that a detached thread created with new
| >HAS to be deleted manually ? Or is this only in case it might still
| >be running?
|
| Firstly, you must create every detached thread using new since it will
| delete itself, literally calling delete this.
| Calling wxThread::Delete() is a correct way to terminate a thread, but
| manually deleting (using delete) a detached wxThread object is not.
| wxThread::Delete() will ask the thread to exit, the thread should check for
| this in wxThread::Entry() regularly using wxThread::TestDestroy() and exit
| when asked to do so.
|
| >(In general I have a unsatisfied felling about when delete is
| >neccessary and when not -- "I only know, it's not , if the class is
| >derived from wxWindows")
|
| For wxThreads: joinable threads must be deleted (when allocated on the
| heap), detached threads may never be deleted. For other classes, consult
| the documentation ;)
|
|
| >Thanks for some feedback,
| >Sebastian
|
| Regards,
|
| Pieter.
+-----------------------
tracking some kind of deadlock bug in Linux.
seems to be in ReadMailcap, src/unix/mimetypes.cpp in wxwindows sources
src/unix/mimetype.cpp:2312
SOLUTION: compile with -pthread on every compile and link line.