convert/doc/getting_serious.qbk
2014-11-22 10:07:46 +11:00

45 lines
2.5 KiB
Plaintext

[/
Copyright (c) Vladimir Batov 2009-2014
Distributed under the Boost Software License, Version 1.0.
See copy at http://www.boost.org/LICENSE_1_0.txt.
]
[section:error_detection Better Error Detection]
[import ../example/getting_serious.cpp]
[getting_serious_example1]
The code above is straightforward and self-explanatory but, unfortunately, strictly speaking, is not entirely deterministic as -1 might be the result of a conversion failure or the successful conversion of the "-1" string. Still, in reality "spare" values are quite often available outside the valid\/sensible range to indicate conversion failures. If so, such deployment might be adequate without introducing non-deterministic behavior. Alternatively, it might be not that uncommon to ignore conversion failures altogether and to simply log the failure and to proceed with the supplied fallback value.
Applications outside these mentioned categories still require conversion failure reliably detected and processed accordingly. The `boost::lexical_cast`'s answer is to throw on failure and ['Boost.Convert] supports that behavior as well:
[getting_serious_example2]
However, to cater for a wider range of program-flow variations, ['Boost.Convert] adds the flexibility of
* delaying the moment when the conversion-failure exception is actually thrown or
* avoiding the exception altogether.
[getting_serious_example3]
Now [@boost:/libs/optional/index.html `boost::optional`] steps forward as the actual type returned by `boost::convert()` which until now we avoided by immediately calling its value-accessor methods:
int i1 = boost::convert<int>(str1, cnv).value();
int i2 = boost::convert<int>(str2, cnv).value_or(fallback_value);
int i3 = boost::convert<int>(str3, cnv).value_or_eval(fallback_function);
Still, from the user perspective, `boost::lexical_cast` processes failure in a somewhat one-dimensional non-negotiable manner. `boost::convert` takes a more flexible approach, provides choice and leaves the decision to the user. It probably is not unimaginable that, on the library level, propagating the conversion-failure exception might be the only sensible response. On the application level though, in my personal experience, the choice has overwhelmingly been to handle conversion failures locally, i.e. avoiding conversion-failure exception propagation or, better still, avoiding exceptions altogether with program flows similar to:
[getting_serious_example4]
and
[getting_serious_example5]
[getting_serious_example6]
[endsect] [/section:error_detection Better Error Detection]