Re: confused by the wierd "q" option of "string.format".

[Ross Berteig <Ross@...>]

At 01:06 PM 7/7/2011, you wrote:
> On 07/07/2011 21.20, tsuui wrote:
> > Let's see the documented example:
> >          string.format('%q', 'a string with "quotes" and \n 
> > new line')
>
> Using %q will translate every char with special meaning in the 
> context of string literals into an escape sequence.
> ....
> gives:
>
> a string with "quotes" and
>  new line
>
> So what Lua returns is perfectly in line with the specs.

There have been cases where I've really disliked this particular 
choice. Both \<newline> and \n are indeed ways to embed a 
newline in a string literal. However, when using %q for 
human-readable log files, the choice it makes breaks up the 
natural formatting of one line per record. In those cases, \n is 
the better choice. Worse, lots of other control characters are 
left intact in the output where they might be lost, missed, or 
damaged by down-stream processes.

For my cases, adding %Q to mean do everything %q does but 
translate every byte outside the range of [\32-\126] to either a 
single-letter escape or to \<digits> (or better, \x<hex>) would 
still preserve the property that the string printed is legal Lua 
while removing any dependency on preserving control characters 
or any damage to log files or other relatively fixed-format uses.

I note that in my Win7 console, Lua 5.1.4 does the following:

    C:...>lua -e "print(string.format('%q','abc\a'))"
    "abc"

It may or may not have emitted the \007 represented by the \a, 
but if so it was visually lost (and it didn't even go 'ding' so 
I've got some configuration details to check into also). 
Checking with GnuWin32's port of od(1) we learn that the \a was 
indeed in the string constant:

    C:...>lua -e "print(string.format('%q','abc\a'))" | od -t x1
    0000000 22 61 62 63 07 22 0d 0a
    0000010

If a human is expected to see the result of the %q, then I 
believe that a transform as I've described for %Q would be 
superior. I'd propose changing %q to do this, but I have no idea 
what unexpected consequences that might have. Adding a new 
format code has the advantage that it can't change the behavior 
of any working program. As a mnemonic, `Q` is obviously stronger 
than `q`, so it even makes sense.

Ross Berteig                               Ross@CheshireEng.com
Cheshire Engineering Corp.           http://www.CheshireEng.com/