mikeash.com pyblog/friday-qa-2013-05-17-lets-build-stringwithformat.html comments

KP - 2018-09-27 02:03:13

Thu, 27 Sep 2018 02:03:13 GMT

else if(next == '@') {
id value = va_arg(arguments, id);
}
results into EXC_BAD_ACCESS see https://screencast.com/t/74BIGsOq2m can you suggest a fix?

sdeitch@robust-sw.com - 2013-06-17 06:34:20

Mon, 17 Jun 2013 06:34:20 GMT

hi,
Very interesting. I am a new reader of your blog and to iOS. I am still trying to build up a appreciation of when to use what approach, hence the following questions..

I am sure that you know that to get the lowest bit you can do intpart & 1 and then shift one place use intpart >>= 1 Since you are "talking" about bits this would be more explicit, was there a reason you did not do this ?

With the writeUnsignedLongLong why did you not just divide by 10, put the results in a temp buffer and place the reverse of the temp buffer in the output?

thanks for your help.

mikeash - 2013-06-05 17:03:15

Wed, 05 Jun 2013 17:03:15 GMT

Yep, I totally agree, and I'm definitely grateful for the additional information.

This code is definitely not intended for real-world use. After all, Apple already provides a better built-in implementation, so just use theirs....

Bruce Dawson - 2013-06-05 16:51:08

Wed, 05 Jun 2013 16:51:08 GMT

I can't argue with the "Let's Build" ethos.

Just beware of the significant challenges in trying to make printing of doubles and floats both efficient and correctly rounded. For my "Let's Build" code for printing floats I decided to ignore both of these, which allowed for much simpler code (and better elucidation) but clearly made my code less generally useful.

mikeash - 2013-06-04 13:14:34

Tue, 04 Jun 2013 13:14:34 GMT

Why not use a pre-made function? Because this is Let's Build, not Let's Find Some Existing Code and Call It. Otherwise the post could be reduced down to a single sentence that says, "Just call sprintf."

Bruce Dawson - 2013-06-03 22:00:21

Mon, 03 Jun 2013 22:00:21 GMT

Why not use David Gay's dtoa() function? It is a well tested standard conversion function. Any homegrown replacement is likely to reinvent many long since fixed bugs, or at least will have inferior rounding.

See http://www.exploringbinary.com/java-hangs-when-converting-2-2250738585072012e-308/ for examples of conversion bugs. The articles on this site also give some suggestions on tricky values to test on.

Technically it is easy to print floats -- I published some straightforward code here:
http://randomascii.wordpress.com/2012/03/08/float-precisionfrom-zero-to-100-digits-2/

This code can easily be extended to doubles. However making the code efficient without breaking it is very hard. Adding rounding is left as an exercise for the reader.

Matt - 2013-05-21 22:30:21

Tue, 21 May 2013 22:30:21 GMT

Mike,
If you're interested in accurate printing/decoding of floating point numbers you should check out Bruce Dawson's blog: Random Ascii - http://randomascii.wordpress.com/category/floating-point/

It's worthing noting that almost all vendor implementations of printf- functionality deviate from the ISO spec in some minor ways, which can create issues with cross-platform code (same inputs != same outputs).

mikeash - 2013-05-18 23:15:04

Sat, 18 May 2013 23:15:04 GMT

Thank you for pointing that out, I've fixed it by moving the negative check to the top, and updated the article.

Jordan Rose - 2013-05-18 05:25:33

Sat, 18 May 2013 05:25:33 GMT

Bug report: writeDouble will print negative infinity as "INFINITY", even though it's distinct from positive infinity.

Gwynne Raskind - 2013-05-17 21:59:07

Fri, 17 May 2013 21:59:07 GMT

Ari: Unfortunately, no. A real -stringWithFormat implementation has to handle, among other things, positional parameters and a wide array of options in each format specifier. There's no support in C for modifying the contents of a va_list on the fly, which would be required for correctly supporting positional parameters when wrapping the existing printf-style functions. The options are 1) using something other than % as your custom specifier character, or 2) reimplementing sprintf from the ground up. Apple did the latter in Core Foundation; look at the function __CFStringAppendFormatCore() in http://opensource.apple.com/source/CF/CF-744.18/CFString.c for an incredibly complicated example :).

Ari Weinstein - 2013-05-17 20:24:40

Fri, 17 May 2013 20:24:40 GMT

Great article.

All of the various implementations of printing out different types is fun, but if you actually wanted to write your own stringWithFormat:, wouldn't you just handle instances of "%@" and then wrap sprintf?