change get_* to accept a prompt as argument

[dmalan]

Instead of re-prompting user in cases of invalid input with Retry:, should instead just re-prompt user with argument. I.e.:

string get_string(string prompt);

Planned for Fall 2017.

[kzidane]

Or maybe get_string(string prompt, string retryPrompt) in case they're different?

[dmalan]

Mm, adds unnecessary complexity I'd say!

[kzidane]

Was just thinking about the scenario where hint may be provided to the user, in case she doesn't co-operate. But yeah, probably unnecessarily complex.

[IvanJasenov]

@dmalan To which function this feature should be applied to? Just get_string or all?
I think it will be good to set number of tries, and if user still makes a mistake, provide him with the appropriate man page for the function at the terminal.

[dmalan]

Oops, sorry for missing this, the vision for this new variant is that it'd behave identically to the current version except that the user would see the prompt 1 or more times ( and instead of Retry:). Though we're not sure yet how we could roll this out seamlessly with CS50 AP still in progress, so no need to forge ahead yet!

[dmalan]

@crossroads1112, @glennholloway, @kzidane:

Inspired by https://gustedt.wordpress.com/2010/06/03/default-arguments-for-c99/, I started working on this, porting the idea to a 0-or-1 argument scenario, https://github.com/cs50/libcs50/blob/optional-args/src/cs50.h#L49-L60.

A function like get_int is then declared as https://github.com/cs50/libcs50/blob/optional-args/src/cs50.h#L123-L125.

This works as intended when compiled with std=c89 and std=gnu99 but not when std=c99 is used, which appears to trigger __STRICT_ANSI__, per http://stackoverflow.com/a/5331540. The result is that compiling the library yields:

$ make build
rm -rf "build"
mkdir -p "build/usr/include" "build/usr/lib" "build/usr/share/man/man3" "build/usr/src"
gcc -c -fPIC -std=c99 -Wall -o "build/cs50.o" "src/cs50.c"
src/cs50.c:224:12: error: too few arguments to function call, single argument 'prompt' was not specified
    return get_int();
           ^~~~~~~~~
src/cs50.h:61:22: note: expanded from macro 'get_int'
#define get_int(...) ZERO_OR_ONE_ARG(get_int, ##__VA_ARGS__)
                     ^               ~~~~~~~
src/cs50.h:60:97: note: expanded from macro 'ZERO_OR_ONE_ARG'
#define ZERO_OR_ONE_ARG(NAME, ...) NAME(_ZERO_OR_ONE_ARG(NAME, ARGS(__VA_ARGS__), ##__VA_ARGS__))
                                   ~~~~                                                         ^
src/cs50.c:196:1: note: 'get_int' declared here
int get_int(string prompt)
^
1 error generated.
make: *** [build] Error 1

even though the macros ensure that, after preprocessing, get_int is indeed called with exactly one argument. Unfortunately, that output doesn't provide a hint (with the usual square brackets) as to how we might suppress that error with a flag.

Thoughts on how best to resolve?

Note that these new get_* functions (e.g., https://github.com/cs50/libcs50/blob/optional-args/src/cs50.c#L199-L234) aren't yet fully implemented. The goal for the coming year is:

• if get_* is called with an explicit prompt, display (and re-display) that prompt as needed
• if get_* is called without an explicit prompt, display no prompt (as in Fall 2016) but re-display Retry: (an implicit prompt) as needed
• if Get* is called without an explicit prompt, behave as in Fall 2014, but display deprecation warning at compilation time
• if Get* is called with an explicit prompt, don't compile

[dmalan]

PS, the goal is only to support these several modes for a year. After Fall 2018, we'd remove Get* altogether and also require an explicit prompt for get_*.

[dmalan]

Actually, this seems to be the result of disparate behavior when you're explicitly in std=c99 mode, per https://gcc.gnu.org/onlinedocs/cpp/Variadic-Macros.html:

... the ‘##’ token paste operator has a special meaning when placed between a comma and a variable argument. If you write

#define eprintf(format, …) fprintf (stderr, format, ##__VA_ARGS__)

and the variable argument is left out when the eprintf macro is used, then the comma before the ‘##’ will be deleted. This does not happen if you pass an empty argument, nor does it happen if the token preceding ‘##’ is anything other than a comma.
...
CPP retains the comma when conforming to a specific C standard. Otherwise the comma is dropped as an extension to the standard.

In particular, if you compile

#define get_int(...) ZERO_OR_ONE_ARG(get_int, ##__VA_ARGS__)
...
int i = get_int();

as foo.c with gcc -E foo.c, you get

int i = ZERO_OR_ONE_ARG(get_int);

as intended. But if you compile the same with gcc -E -std=c99 foo.c, you get

int i = ZERO_OR_ONE_ARG(get_int,);

with the undesired extra comma. The result is that the other macros, https://github.com/cs50/libcs50/blob/optional-args/src/cs50.h#L54-L60, no longer pattern-match as intended.

Thoughts on thus changing the Makefile not to specify std explicitly? What might break on wha platforms?

[dmalan]

Discussed in person.