Misc. minor fixes

[Daniel-Cortez]

What this PR does / why we need it:

Fixes for various minor bugs that weren't worth making a separate PR for each of them:

• Fixed invalid formula for calculating array size (in characters) in function hier1().
  The cell size in bits was hardcoded as 32 (should be sizeof(cell)*8), which could lead to generation of invalid code for array boundary checks if the Pawn toolkit was configured with 16- or 64-bit cell size.
  Of course, this bug doesn't affect SA-MP (which is 32-bit), but I think it's still good to have this fix here.

• Fixed invalid type cast in funcstub().
  See refactoring of the codebase #296 (comment).

• Fixed OOB array access in encode_cell().
  This one seems to originate from a commit that I made a year and a half ago (sorry!)

• Fixed memory leaks in plungequalifiedfile().
  Special thanks goes to @YashasSamaga for reporting the bug, as well as providing a fix, though I slightly changed that fix in the following ways:

  • Moved free(path) before error(103), because path won't be freed otherwise. This is because on fatal errors (1xx) error() returns (via longjmp()) directly to an error buffer set in pc_compile, so the code after such error() call is never executed.
  • For the same reason I removed two returns, as they are pretty much useless after error(<fatal error code>).

• Removed NULL pointer checks before the uses of function free().
  free() does nothing if the pointer is NULL, so any NULL checks before the use of free() are redundant.

• Fixed invalid uses of realloc() in several places.
  The old pointer could have got leaked if the function failed to resize the old memory block or allocate a new one.

• Simplified the code in function pushstk() by replacing a group of calls to malloc()/memcpy()/free() with one realloc().

• Removed function do_switch() (sc6.c).
  Apparently this function was a duplicate of do_jump(); it didn't implement any extra checks, so its only difference from do_jump() was in its name.

• Simplified string emptiness checks.
  Using strlen(<string>)==0 to identify empty strings is inefficient because if the string is not empty strlen() traverses through the whole string to get its length.

• Fixed potential NULL pointer dereferences in functions sc_path() and setconfig().

• Fixed incorrect error message when compact encoding is ineffective (Incorrect fatal error when compact encoding is ineffective #425).

• Other minor adjustments.

Which issue(s) this PR fixes:

Fixes #425

What kind of pull this is:

• A Bug Fix
• A New Feature
• Some repository meta (documentation, etc)
• Other

Additional Documentation:

[Y-Less]

Looks fine to me, but only from a visual readthrough.

  * For the same reason I removed two `return`s, as they are pretty much useless after `error(<fatal error code>)`.

I couldn't find those returns. Did you change this part? I'd suggest not removing returns after fatal errors, since the c compiler doesn't know that error might not return (it could be marked as noreturn, but it does in some circumstances). Thus the return will cut off certain code paths and allow the compiler compiler to do more when compiling the compiler.

[Daniel-Cortez]

I couldn't find those returns. Did you change this part?

No, I meant that I removed them from the code @YashasSamaga added in their original fix.

I'd suggest not removing returns after fatal errors, since the c compiler doesn't know that error might not return (it could be marked as noreturn, but it does in some circumstances). Thus the return will cut off certain code paths and allow the compiler compiler to do more when compiling the compiler.

Makes sense, but I removed those returns not only because they would be unreachable, but also to be more in-line with the existing code (in all the other code there are usually no returns after fatal errors).

[Y-Less]

Oh I see, then fair enough.

[Zeex]

Can we have the empty string check defined as a macro please? it would be more readable IMO

[Daniel-Cortez]

@Zeex Done.

[Zeex]

Thanks, looks good