|
libcfg+ - Viewing Bug Advanced Details |
[ Go Back ]
|
|
|
ID:
|
Category:
|
Severity:
|
Reproducibility:
|
Date Submitted:
|
Last Update:
|
|
39 |
library core |
minor |
always |
2003-02-24 23:10 |
2003-08-20 13:33 |
|
|
Reporter:
|
rajo |
Platform:
|
|
|
|
Assigned To:
|
|
OS:
|
|
|
|
Priority:
|
normal |
Version:
|
|
|
|
Status:
|
feedback |
Product Version:
|
current from CVS |
|
|
Product Build:
|
|
Resolution:
|
open |
|
|
Projection:
|
none |
Duplicate ID:
|
|
|
|
ETA:
|
none |
|
|
|
Summary:
|
Integers cannot be initialized in float-like style |
|
Description:
|
Integers, long integers and similar signed/unsigned datatypes cannot be initialized in float-like style. Float like style is something like 10e23, 1.6e9 and so on. |
|
Steps To Reproduce:
|
Create integer option. Try to initialize with --integer=10e23 on command line. |
|
Additional Information / Solution:
|
Parsing sequence:
1. Fetch string.
2. Try to convert (normal conversion).
3. If OK, go [6].
4. If failed try to conver with float conversion.
5. If OK go [8].
6. Return with CFG_ERR_BADNUMBER.
7. If there are decimal numbers return with CFG_ERR_BADNUMBER.
8. Check range and do rest of the stuff in old way. |
|
Attached Files:
|
|
|
Bug Notes |
|
|
|
I started with this issue implementation and come to one problem. I do not know how to detect if double/float value has numbers beyond the decimal point. This is current part of source:
<code>
if (double_val - floor(double_val) != 0.0)
return CFG_ERROR_BADNUMBER;
if (type == CFG_ULONG) {
ulong_val = (unsigned long) double_val;
} else {
long_val = (long) double_val;
}
</code>
However, the code above does the job and perfectly work. The only problem is, that there is floor() function used. If you want to use floor() or another function from math.h, you have to link your application or library with -lm switch, which forces linking against mathematical library (libm).
Basically, I want to get rid of mathematical library usage. It seems to be contraproductive to link whole stuff against math library due to only one call. Therefor we need to discover other way how to detect if there are numbers beyond decimal point in double number (double_val). Maybe with some type conversion and/or similar approach without math library usage. Ideas and comments are welcome and appreciated. |
|
|
|
|
OK, I fixed the issue. Current code looks like this and seems to work OK:
<code>
if (type == CFG_ULONG) {
ulong_val = (unsigned long) double_val;
if (double_val - (double) ulong_val != 0.0)
return CFG_ERROR_BADNUMBER;
} else {
long_val = (long) double_val;
if (double_val - (double) long_val != 0.0)
return CFG_ERROR_BADNUMBER;
}
</code>
However, we also need to determine error codes to return. There are two relevant:
CFG_ERR_BADNUMBER - cannot convert number to appropriate type
CFG_ERR_OVERFLOW - number conversion overflow
What to return when initializing integer in float-like style and number have decimal characters? (for example: --int=12e-2)
What to return when initializing integer in float-like style, number does not have decimal characters, but it is very big (as big as it cannot be converted even to double)? (for example: --int=12e100000000 <-- very big exponent)
We have to solve questions above. |
|
|
|
|
- -int=12e-2 returns CFG_ERR_BADNUMBER because number is not integer
--int=12e100000000 returns CFG_ERR_OVERFLOW because number is to big (ofcourse for all default data types. ;) |
|
|
|
|
The still persisting problem is, that this behaviour is not completelly conformed.
GOOD:
ondrej@Apolon:~/prog/libcfg+/src$ ./test --int=12e-2
***** argument '12e-2' for option 'int' could not be converted to appropriate numeric type at position 2 on command line
***** cmdline error -5 at 1: option could not be converted to appropriate numeric type
***** opt = 'int', arg = '12e-2', idx = 1, argv[idx] = '--int=12e-2'
BAD:
ondrej@Apolon:~/prog/libcfg+/src$ ./test --int=12e100000000
***** argument '12e100000000' for option 'int' could not be converted to appropriate numeric type at position 2 on command line
***** cmdline error -5 at 1: option could not be converted to appropriate numeric type
***** opt = 'int', arg = '12e100000000', idx = 1, argv[idx] = '--int=12e100000000'
Both returns the same error code (-5), what is CFG_ERR_BADNUMBER. |
|
|
|
|
This was fixed in shared.c, revision 1.34.
However there is still this problem:
If too big decimal number occurs, CFG_ERR_OVERFLOW is returned, because we do not have any solution how to detect decimal numbers for very big numbers. It this OK? |
|