Code fix / improvements #60

ledvinap · 2019-05-13T19:04:31Z

Apply some changes to existing code. Split into separate commits with comments

Handle 0x / 0b when width is equal to string size (issue mpaland#50): printf("%#4x", 0x1234); -> "0x1234" Do not print # prefix if it does not fit into PRINTF_NTOA_BUFFER_SIZE %#<prec>o is implemented as in libc: - printf("%#0o", 0); does output octal prefix ("0") - printf("%#3o", 1); printf "001" - padding zero is used as octal prefix (issue mpaland#53) Left padding and precision is handled correctly (issue mpaland#49): - printf("%-10.6d", 1024); -> "001024 "

Fixes issue mpaland#51 - fractional part may overflow in round-to-even part Rounding is still inexact, some numbers are rounded wrong way

printf("%#.0f", 1) -> "1."

Old code did use -308 as exponent for 0.0, now value is printed with zero exponent

- %g precision is number of significant digits. Default %g precision shall be honored when printing as %f printf("%g", 123.4567); -> "123.457" - %g prints zero correctly (except for trailing zeroes) printf("%g", 0); -> "0.0" - %g shall switch to %f if supplied value fits into specified precision printf("%.8g", 12345678); -> "12345678"

Negative value passed by "%.*" shall be taken as if no precision was specified

uses _out_pad, removes some unnecessary tests (length of string to print is known)

Signed overflow is undefined in C, this was causing problem when printing INT_MIN (gcc sign-extended value first, so 2^64-INT_MIN got printed)

codecov-io · 2019-05-13T19:10:56Z

Codecov Report

Merging #60 into master will decrease coverage by 1.37%.
The diff coverage is 93.82%.

@@            Coverage Diff             @@
##           master      #60      +/-   ##
==========================================
- Coverage     100%   98.62%   -1.38%     
==========================================
  Files           1        1              
  Lines         359      363       +4     
==========================================
- Hits          359      358       -1     
- Misses          0        5       +5

Impacted Files	Coverage Δ
printf.c	`98.62% <93.82%> (-1.38%)`	⬇️

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 21eea6c...e29f0f8. Read the comment docs.

axiomlegend · 2019-05-17T19:28:25Z

Calling *out for each character is costly. Consider sprintf functionality. you call *out for every character only to do: if (a < b) a[i] = c; Lot of overhead for a few instructions.
Definitely need to put most options to ntoa_format, etc. into a struct vs stack.

ledvinap · 2019-05-18T13:10:51Z

@axiomlegend :
Yes, out is quite expensive, but also generic. In a lot of embedded applications output to serial port is (or SWO) desired, and using sprintf will complicate things a lot.
One improvement is to use write semantics for output. Number of calls will be reduced, 1 write between %, 1 call for value output and one (or few chunks) for padding.
Another optimization in my list is to move buffer limit test out of out function, slightly improving performance.

ledvinap added 11 commits May 13, 2019 19:41

Replace padding loops with function

87d27ef

Fix/improve %f rounding code

2706024

Fixes issue mpaland#51 - fractional part may overflow in round-to-even part Rounding is still inexact, some numbers are rounded wrong way

Implement %#.0f

4163121

printf("%#.0f", 1) -> "1."

Minor cleanup of _ftoa code

9a40aa2

Print 0 as 0.0 / 0.0e+00

c6b9866

Old code did use -308 as exponent for 0.0, now value is printed with zero exponent

Simplify flags handling a bit

e88e741

Handle negaive precision correctly

f511603

Negative value passed by "%.*" shall be taken as if no precision was specified

Simplify %s a bit

184e14a

uses _out_pad, removes some unnecessary tests (length of string to print is known)

Fix signed value overflow

e29f0f8

Signed overflow is undefined in C, this was causing problem when printing INT_MIN (gcc sign-extended value first, so 2^64-INT_MIN got printed)

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Code fix / improvements #60

Code fix / improvements #60

ledvinap commented May 13, 2019

codecov-io commented May 13, 2019

axiomlegend commented May 17, 2019

ledvinap commented May 18, 2019

Code fix / improvements #60

Are you sure you want to change the base?

Code fix / improvements #60

Conversation

ledvinap commented May 13, 2019

codecov-io commented May 13, 2019

Codecov Report

axiomlegend commented May 17, 2019

ledvinap commented May 18, 2019