Failing to emit line directive if included file has a #if or #define on the first line

[njnobles]

Wave is not emitting the line directive if the first line in the included file is a #if or #ifdef that evaluates true and has content in its block. Likewise, it won't emit the line directive if the first line is a #define. This issue only occurs if there are not multiple #define or #if occurring on subsequent lines.
When using default_preprocessing_hooks, the above applies if the #if/#ifdef/#define is the first non-empty line or non-comment line. Essentially, if the directive immediately precedes the first content emitted from the included file. With eat_whitespace, the directive must be on the very first line of the include file to exhibit the issue.

Not sure if I've explained it very well, but I think an example will help better demonstrate the issue.

Example:
a.cpp

#include "if.h"
#include "def.h"
test

if.h

#if 1 //This must be on the first line for eat_whitespace, can be on 2nd, 3rd etc, for default_preprocessing_hooks as long as it's whitespace before it
if_h
#endif

def.h

#define A
def_h

Resulting preprocessed file:
a.preproc

if_h

def_h

Expected output:

#line 2 "if.h"
if_h
#line 2 "def.h"
def_h

As mentioned, the line directives do get emitted correctly if the include file is something like this:

#define A
#define B
content

So it's only occurring when there's a single directive at the beginning.

[jefftrull]

Thank you for finding this excellent corner case!

I'm still exploring the issue but I find it's actually more general: any blank line (or in your case, a directive that becomes a blank line) will cause the line directive to be suppressed. For example:

a.cpp

#include "b.h"

b.h

B

Running Wave on a.cpp produces:

B

i.e., no line directive showing that B came from b.h appears.

I think this is because the logic that determines whether to emit a line directive checks only whether the number of newlines (as perceived by the reader of the preprocessed code) is correct, and not whether the filename has changed.

[jefftrull]

Confirmed! Thanks for the report.