We are running a survey to help us improve the experience for all of our members. If you see the survey appear, please take the time to tell us about your experience if you can.
I am trying to convert an unmaintainable user interface driver over to something that uses data structures to define the menu organization (This example is stripped down to the bare minimum from my actual app).
Picture an array of screens, selectable by the user's activation of some arrow keys:
Col 0 Col 1 +-----------+ Row 0 | Screen 1a | +-----------+ /\ || \/ +-----------+ +-----------+ Row 1 | Screen 2a | <===> | Screen 2b | +-----------+ +-----------+
I define a function to output each screen (in the code below, they also change the row & column indices):
void f1a(void) { outputScreen1a(); if(r > 0) r++; return; } /* ============= */ void f2a(void) { outputScreen2a(); if(c < 1) c++; return; } /* ============= */ void f2b(void) { outputScreen2b(); r = c = 0; return; }
These are called by the main() function, using a function pointer table:
static code const void (* code menuFunc[2][2])(void) = { { f1a, NULL }, { f2a, f2b } }; unsigned char r, c; void main(void) { unsigned char i; SFRPAGE = 0; /* Go to default page */ WDTCN = 0xDE; /* Disable watchdog timer */ WDTCN = 0xAD; IE = 0; /* Disable all interrupts */ i = 0; r = c = 0; while (1) { if(i < 70) { i++; } else { i = 0; (*menuFunc[r][c])(); } } /* END while (TRUE) */ } /* END main() */
As presented above, the source file will compile without error. But my application is quite a bit more complex than what I present here, the table is assymetrical (marked by the NULL pointer) and I want to guantee I do not inadvertently dereference it, so I add a qualification to the main control loop:
while (1) { if(i < 70) { i++; } else { i = 0; if((menuFunc[r][c])() != NULL) (*menuFunc[r][c])(); } } /* END while (TRUE) */
C51 does not like this. It flags the statement where I check for NULL:
?????.C:(56): error C193: '!=' : illegal conversion from/to 'void'
NULL is defined in stdio.h:
#define NULL ((void *) 0)
K&R doesn't state it explicitly, but seems to assume that void is an object type, that (void *) can only point to an object, and never uses NULL as a pointer to a function.
So the question is, if I can't use NULL to mark an uncallable function in the table, is there a null function pointer I could use instead?
============================================================ Gary Lynch | To send mail, no$pam in domain name lynchg@no$pam.com | must be changed to stacoenergy. ============================================================
I don't own PC-Lint.
I wrote a little program that extracts a bit of information from the Keil project files (and also looks at the produced output files).
Then I use this to create a Makefile that uses a couple of free compilers to compile (but not link) the source code with suitable warning levels. In my case, it is enough to know what source files are part of the project, and what directories the source is in and what directories to scan for include files.
Not perfect, but a quick way to get a bit of extra static analysis of the code. Spending money on PC-Lint would probably be a better solution, but I did the above once when I was bored and had to wait a couple of hours for a transport. In the end, the solution has produced a couple of intersting warnings that I would have liked support for by the ARM compiler.
Alas, it can't do cross-module analysis, so I should probably some day look at automatic support for any of the free source-code analyzers that are available. When I originally did the above, I didn't had Internet access, so I had to settle for whatever tools/info I had on the laptop.
Anyone who have any special favourites? en.wikipedia.org/.../List_of_tools_for_static_code_analysis
I tried splint yesterday. It is a nice tool, but lacks support for specific compiler extensions besides some GNU stuff.
What I forgot to mention is that I do the above for ARM and not C51 projects. This greatly reduces the problems with compiler extensions.
A bit of creative use of sed or #defines can normally "standardize" the code.
On 24-Oct-2008 05:50; Per Westermark wrote: > > I wrote a little program that extracts a bit of information > from the Keil project files (and also looks at the produced > output files). > > Then I use this to create a Makefile that uses a couple of > free compilers to compile (but not link) the source code > with suitable warning levels. > I keep digressing, but I have been trying to implement just what you describe for years.
I have batch files that call Perl scripts to extract enough info from the *.__i and *.lnp files to create a makefile I could archive with my source. It would need to include the compile- & link-time directives (which I use for configuration management). I can create them for a specific project, but have never gotten one to work as a general purpose utility.
- What language do you write in? - Did you get yours working for the general case?
I really have to check if I used C or C++ ;)
Not the quickest way to write the string manipulation but I didn't have too much tools installed on the laptop, and I got the idea when I failed to buy a book to read while waiting for a plane after the original plane got cancelled.
Yes, it worked quite well, but in this case I wasn't originally so very concerned about compilation options since the goal was to test-compile with other compilers. The important thing was to be able to feed the other compilers with define symbols. My aim was to use the correct set of source files, the correct include directories and the correct conditional compilation options. Then I had to do a bit of automatic tweaking of the source to get it through for example the MinGW version of the gcc compiler.
But yes, it does pick up the command-line options that the ARM compiler prints at the top of the list files, so I can compile the files with the ARM compiler too and with the correct options.
I basically did this instead of doing cross words, without really knowing what would come out of it.