implement yyset_interactive

[rmccullagh]

Currently, only available in reentrant mode, since it needs a scanner argument. However, I believe a yyinteractive macro can be implemented too. References #290

[Explorer09]

When you are implementing yyset_interactive, could you also remove this yy_set_interactive macro, which could cause confusion.

[rmccullagh]

@Explorer09 Good call, I will get this updated tonight.

[rmccullagh]

@Explorer09 This is done.

[jannick0]

Sorry for my ignorance: Why not simply amending the definition of the macro yy_set_interactive using the m4 variables M4_YY_PROTO_LAST_ARG and M4_YY_DECL_GUTS_VAR already defined in flex.skl?

[EDIT] ... like so?

--- C:/myDevel/myprogs/flex/flex_2.6.4/yy_interactive/flex1.skl	Wed Aug 23 13:00:27 2017
+++ C:/myDevel/myprogs/flex/flex_2.6.4/yy_interactive/flex.skl	Tue Dec 12 17:19:38 2017
@@ -611,8 +611,9 @@
 m4_ifdef( [[M4_YY_NOT_IN_HEADER]],
 [[
 #define yy_new_buffer yy_create_buffer
-#define yy_set_interactive(is_interactive) \
+#define yy_set_interactive(is_interactive M4_YY_CALL_LAST_ARG) \
 	{ \
+    M4_YY_DECL_GUTS_VAR(); \
 	if ( ! YY_CURRENT_BUFFER ){ \
         yyensure_buffer_stack (M4_YY_CALL_ONLY_ARG); \
 		YY_CURRENT_BUFFER_LVALUE =    \

Are there any issues when doing so? Of course, I see the advantage of having a function instead of a macro, especially from another c file, however, macros are preferred by flex.

Separately, would it be fair to say that the remainder of the changes are enhancements to command line options only?

[Explorer09]

@jannick0 The use of the function instead of a macro is because YY_CURRENT_BUFFER_LVALUE is an internal macro, AFAIK. That won't be the problem if yy_set_interactive is called only within the .l file, but that's not the case.

[jannick0]

@Explorer09 Agree, but the issue then becomes a philosophical one, namely if the implementation should prefer macros (as is essentially right now) or functions. Using a wrapper would a way out of the current situation to effectively call macro defined in the lex file only without any changes.

The short patch above is amending what I would consider a bug, since for reentrant scanners the opaque scanner object needs to be resolved as it contains the flag is_interactive as a member to be set.

[rmccullagh]

@jannick0 If you read #290, you can see that yy_set_interactive is perfectly callable from the .l file, but if you need more control, you don't have access to that macro because it is not defined in a header file. Furthermore, the required data structure to implement yy_set_interactive is an internal data structure.

[Explorer09]

There are weaknesses of C macro mainly because they work like simple string substitution. If you only use C macro for reasons of function inlining, please don't, or do it sparingly.

I should have mentioned that yy_set_interactive, in it's current macro form, lack the proper do {...} while (0) enclosure. See this (https://stackoverflow.com/questions/257418/do-while-0-what-is-it-good-for) for reasons.
Above is one of the reasons you should avoid macro as much as possible. Modern compilers are excellent at inlining functions, so they can inline static functions when that benefits. Speaking of static functions, I would like to bring issue #277 together with this, since I would like if there's a way to make yyset_interactive static when we have the function implemented.

[jannick0]

Again, using macros or functions is a question of coding style which is not my business here, supposedly rather the package maintainer's. Other issues about how to use macros are interesting, but to be dealt with in a different place I would say. Of course, I agree that they should certainly be followed for implemented macros.

If a function implementation of yy_set_interactive is ultimately chosen, then I would strongly recommend not defining it as static sucht that it can be made available in other entities. [BTW to avoid any confusion: Even the macro implementation of any procedure like yy_set_interactive (not defined in the flex header file) can be made available in other entities using a public function wrapper defined in the .l file as long as the opaque scanner object type is available in that very entity. The latter is needed either way, the macro of the function implementation.]

[westes]

Could you rebase and resolve conflicts? As you can see, we're (finally) clearing out the backlog.