Re: [Flex-devel] #include <unistd.h> in middle of scanner: bad idea

[Kang-Che S. <exp...@gm...>]

On Thu, Aug 6, 2020 at 1:37 AM Kaz Kylheku <938...@ky...> wrote:
>
> Flex generates a file which contains
> a #include <unistd.h> long after user material. This seems to be for
> the sake of having a declaration of isatty.
>
> The problem with this is that the user code can #define macros that
> can break the header. C programs have to be careful with what they
> do before system headers are included.

System headers can always be overridden by user code. A prominent example is
feature test macros, which are defined before the inclusion of system headers
and can change the ABI used for standard library functions.

> Here is small repro test case:
>
>    %{
>
>    #define pause blah[]
>
>    %}
>
>    %%
>
> Result:
>
>    $ flex lex.l ; cc lex.yy.c
>    lex.l:3:15: error: declaration of ‘blah’ as array of functions
>    #define pause blah[]

Why would anyone want to do that? Redefining a standard library symbol makes
the code unportable (except for local scope redefines).

> How about just an "extern int isatty(int)" somewhere?

Forward declaration can somewhat defeat the purpose of the header. A library
implementation might not declare the function exactly as described in the
standard document. It might contain other attributes or ABI keywords.