According to the online specs, the lexer tries to tokenize by maximal matching (except for one exception in the case of ranges like "1..2"). The fact that this exception is stated seems to indicate that it's permitted to have two literals side-by-side without an intervening space.

So does that mean "1e2" should be tokenized as (float lit: 1e2) and
"1f2" should be tokenized as (int lit: 1)(identifier: f2)?

Or, for that matter, "123abcdefg" should be tokenized as (int lit:
123)(identifier: abcdefg) whereas "0x123abcdefg" should be tokenized as
(int lit: 0x123abcdef)(identifier: g)?

Or worse, if we still allow octals, "0129" should be tokenized as (octal
lit: 012)(int lit: 9)?

Or do we expect that any integer/float literal will always span the longest string that has characters permitted in any numerical literal, and then after the fact the lexer will give an error if the string cannot be interpreted as a legal literal? IOW, "0129" will first be scanned in its entirety as a numerical literal, then afterwards the lexer decides that '9' doesn't belong in an octal so it throws an error (as opposed to maximally matching "012" as an octal literal followed by a decimal literal "9").  Or, for that matter, "0123xel.u123" will be scanned as a numerical literal (since all the characters in it occur in some kind of numerical literal), and then an error generated after the fact when the lexer realizes that this string isn't a legal numerical literal?


T

-- 
All men are mortal. Socrates is mortal. Therefore all men are Socrates.

On 02/11/2012 07:42 PM, H. S. Teoh wrote:
> According to the online specs, the lexer tries to tokenize by maximal
> matching (except for one exception in the case of ranges like "1..2").
> The fact that this exception is stated seems to indicate that it's
> permitted to have two literals side-by-side without an intervening
> space.
>
> So does that mean "1e2" should be tokenized as (float lit: 1e2) and

Yes.

> "1f2" should be tokenized as (int lit: 1)(identifier: f2)?
>

No. maximal munch:

(float lit: 1f)(int lit 2)


> Or, for that matter, "123abcdefg" should be tokenized as (int lit:
> 123)(identifier: abcdefg)

Yes.

> whereas "0x123abcdefg" should be tokenized as
> (int lit: 0x123abcdef)(identifier: g)?
>
> Or worse, if we still allow octals, "0129" should be tokenized as (octal
> lit: 012)(int lit: 9)?
>

DMD views 0129 as an error. Therefore, the best way to handle integer literals with initial 0 is to just parse them as decimal and to reject them if they exceed 7.

> Or do we expect that any integer/float literal will always span the
> longest string that has characters permitted in any numerical literal,
> and then after the fact the lexer will give an error if the string
> cannot be interpreted as a legal literal? IOW, "0129" will first be
> scanned in its entirety as a numerical literal, then afterwards the
> lexer decides that '9' doesn't belong in an octal so it throws an error
> (as opposed to maximally matching "012" as an octal literal followed by
> a decimal literal "9").  Or, for that matter, "0123xel.u123" will be


(int lit: 0123)(identifier: xel)(token: '.')(identifier: u123)

> scanned as a numerical literal (since all the characters in it occur in
> some kind of numerical literal), and then an error generated after the
> fact when the lexer realizes that this string isn't a legal numerical
> literal?
>
>
> T
>

No. As an example, that kind of processing the code would reject the valid token q{0123xel.u123}.

Just wanted to point you to my working D lexer (needs a CTFE bugfix http://d.puremagic.com/issues/show_bug.cgi?id=6815).

https://gist.github.com/1262321 D part
https://gist.github.com/1255439 Generic part

On Sat, Feb 11, 2012 at 09:59:06PM +0100, Martin Nowak wrote:
> Just wanted to point you to my working D lexer (needs a CTFE bugfix http://d.puremagic.com/issues/show_bug.cgi?id=6815).
> 
> https://gist.github.com/1262321 D part https://gist.github.com/1255439 Generic part

Cool, thanks!

Looks like you've gone far beyond what I'm doing. :-) But it's still a good learning exercise for me to get comfortable with coding in D.


T

-- 
"The whole problem with the world is that fools and fanatics are always
so certain of themselves, but wiser people so full of doubts." --
Bertrand Russell.
"How come he didn't put 'I think' at the end of it?" -- Anonymous

On 02/11/2012 09:59 PM, Martin Nowak wrote:
> Just wanted to point you to my working D lexer (needs a CTFE bugfix
> http://d.puremagic.com/issues/show_bug.cgi?id=6815).
>

This seems to do the job:
constfold.c:1566
-        if (tn->ty == Tchar || tn->ty == Twchar || tn->ty == Tdchar)
+        if (tn->isImmutable() && (tn->ty == Tchar || tn->ty == Twchar || tn->ty == Tdchar))

However, I don't know the compiler's internals at all, therefore it is quite possible that the fix is incorrect.


> https://gist.github.com/1262321 D part
> https://gist.github.com/1255439 Generic part

Bug: The lexer cannot handle /++/ and /**/ (without new line character at the end).

On 02/12/2012 12:35 AM, Timon Gehr wrote:
> On 02/11/2012 09:59 PM, Martin Nowak wrote:
>> Just wanted to point you to my working D lexer (needs a CTFE bugfix
>> http://d.puremagic.com/issues/show_bug.cgi?id=6815).
>>
>
> This seems to do the job:
> constfold.c:1566
> - if (tn->ty == Tchar || tn->ty == Twchar || tn->ty == Tdchar)
> + if (tn->isImmutable() && (tn->ty == Tchar || tn->ty == Twchar ||
> tn->ty == Tdchar))
>
> However, I don't know the compiler's internals at all, therefore it is
> quite possible that the fix is incorrect.
>
>
>> https://gist.github.com/1262321 D part
>> https://gist.github.com/1255439 Generic part
>
> Bug: The lexer cannot handle /++/ and /**/ (without new line character
> at the end).

Another thing.. Using /+ and +/ in strings gives unexpected results when commented out:
/+
auto a = "/+";
+/
everything from this point is commented out.

/+
auto a = "+/";
+/ // already terminated by the string value.

Is this a bug, or as designed? /++/ is meant to comment out code, so it would have been nice if it was able to handle this, but I guess it would complicate the lexer a great deal.

On Sun, Feb 12, 2012 at 01:00:07AM +0100, simendsjo wrote: [...]
> Another thing.. Using /+ and +/ in strings gives unexpected results
> when commented out:
> /+
> auto a = "/+";
> +/
> everything from this point is commented out.
> 
> /+
> auto a = "+/";
> +/ // already terminated by the string value.
> 
> Is this a bug, or as designed? /++/ is meant to comment out code, so it would have been nice if it was able to handle this, but I guess it would complicate the lexer a great deal.

It's designed. At least according to the online specs:

	The contents of strings and comments are not tokenized.
	Consequently, comment openings occurring within a string do not
	begin a comment, and string delimiters within a comment do not
	affect the recognition of comment closings and nested "/+"
	comment openings. With the exception of "/+" occurring within a
	"/+" comment, comment openings within a comment are ignored.

		a = /+ // +/ 1;    // parses as if 'a = 1;'
		a = /+ "+/" +/ 1"; // parses as if 'a = " +/ 1";'
		a = /+ /* +/ */ 3; // parses as if 'a = */ 3;'

For commenting out code, a much better way is to use version(none){...}.


T

-- 
No! I'm not in denial!

On Sunday, February 12, 2012 01:00:07 simendsjo wrote:
> Another thing.. Using /+ and +/ in strings gives unexpected results when
> commented out:
> /+
> auto a = "/+";
> +/
> everything from this point is commented out.
> 
> /+
> auto a = "+/";
> +/ // already terminated by the string value.
> 
> Is this a bug, or as designed? /++/ is meant to comment out code, so it would have been nice if it was able to handle this, but I guess it would complicate the lexer a great deal.

It's by design. Everything between /+ and +/ is a comment. It doesn't matter what it is. There's nothing special about " which would make it ignore the characters following it when looking for the +/ to end the comment.

- Jonathan M Davis

On Sun, 12 Feb 2012 01:00:07 +0100, simendsjo <simendsjo@gmail.com> wrote:

> On 02/12/2012 12:35 AM, Timon Gehr wrote:
>> On 02/11/2012 09:59 PM, Martin Nowak wrote:
>>> Just wanted to point you to my working D lexer (needs a CTFE bugfix
>>> http://d.puremagic.com/issues/show_bug.cgi?id=6815).
>>>
>>
>> This seems to do the job:
>> constfold.c:1566
>> - if (tn->ty == Tchar || tn->ty == Twchar || tn->ty == Tdchar)
>> + if (tn->isImmutable() && (tn->ty == Tchar || tn->ty == Twchar ||
>> tn->ty == Tdchar))
>>
>> However, I don't know the compiler's internals at all, therefore it is
>> quite possible that the fix is incorrect.
>>
>>
>>> https://gist.github.com/1262321 D part
>>> https://gist.github.com/1255439 Generic part
>>
>> Bug: The lexer cannot handle /++/ and /**/ (without new line character
>> at the end).
>
> Another thing.. Using /+ and +/ in strings gives unexpected results when commented out:
> /+
> auto a = "/+";

/+ comments do nest. So you have opened two levels and the comment stops after two pairing +/.
/* comments do not nest.

> +/
> everything from this point is commented out.
>
> /+
> auto a = "+/";
> +/ // already terminated by the string value.
>
> Is this a bug, or as designed? /++/ is meant to comment out code, so it would have been nice if it was able to handle this, but I guess it would complicate the lexer a great deal.

"Martin Nowak" <dawg@dawgfoto.de> wrote:
> Just wanted to point you to my working D lexer (needs a CTFE bugfix http://d.puremagic.com/issues/show_bug.cgi?id=6815).
>
> https://gist.github.com/1262321 D part https://gist.github.com/1255439 Generic part

Hi, how it should be compiled? I tried with DMD 2.057:
>dmd dlexer.d
and got
>c:\Programs\Programming\Lang\dmd2\windows\bin\..\..\src\phobos\std\conv.d(94): Error: template instance std.format.formatValue!(Appender!(string),defineToken,char) recursive expansion