Thread: A C program to parse a C program

I'm giving myself an exercise at writing a basic parser for reading C programs. My logic involves state switches (in_string, inside_parens, in_block_comment, etc.) and looking at individual characters, as I'm using fgetc() for input.

I'm asking about the proper approach for parsing the file. I've written several hacked-out parsers for various needs over the years, but those were mostly short and sweet and disposable. My objective for writing a parser this time is to isolate all the function calls, standalone or nested.

I usually take a divide and conquer approach, removing the low hanging fruit that I don't care about, like comments, and the contents of strings and trailing blanks, etc.

Once I start reading characters, I'm setting states and keeping stats. How many characters read, how many "words" read (which, depending on your definition of a word, could mean many things), lines read, blah blah.

Once i find an interesting character, like a double quote, or a single quote, or a back slash inside a pair of aposts, I change states.

Should my parser loop, as it progresses with each new character, consider the state I'm in over the character I read, or should the character drive the state logic?

For example

Code:

while(get a char) { 
   if (in_string) { .... }

or

Code:

while (get a char) { 
   if (c=='"') { .... }

What's your opinion?

I'm finding it convenient to keep track of the previous character. I think it would also be handy to keep track of the next character. For example...

previous is '/'
current is '*'
next is '/'

Right now, if I'm working with current, an asterisk, I can see that previous is a slash. If I'm not in a string, then this is the start of a block comment. When I advance to the next loop, previous is now '*' and current is '/', and if I take into consideration current and previous, then I'm obviously at the end of a block comment, but that's not the real case. So, having 'next" would be handy in this case.

Y'all can ignore this. I've decided that for the purposes of just finding all the function calls, I don't need this level of parsing (tokenizing). I can probably just set up a regex to find function names once I de-comment it and isolate the strings, and I have both of those working already.