Hi everybody!
All BCMath functions expects strings as operands (in the strict sense;
do not confuse with arguments). This of course can result in unexpected
behavior if floats are passed, namely:
- floats may be converted to a string in scientific notation
- floats may be converted with a decimal separator which is not a point
Both result in BCMath treating such strings as '0'. Over the years
there have been at least six bug reports regarding this issue: #10116,
#42871, #47633, #55160, #66745, #66959. The last one has been
reclassified as feature request, and I would like to address it.
I already have proposed to introduce bcinit() which would take a string,
int or float, and return a properly formatted string with the given or
default scale, see https://github.com/php/php-src/pull/2746. Nikita
suggested to accept the arguments allowed by bcinit() for all
functions in the first place. This would, however, raise the question
which scale should be applied to floats.
Before proceeding to an RFC (if necessary at all), I'm looking forward
to hear your thoughts regarding this topic.
--
Christoph M. Becker
I already have proposed to introduce bcinit() which would take a string,
int or float, and return a properly formatted string with the given or
default scale, see https://github.com/php/php-src/pull/2746. Nikita
suggested to accept the arguments allowed bybcinit()for all
functions in the first place. This would, however, raise the question
which scale should be applied to floats.
I agree with Nikita. Given that current behavior for floats passed to
bc*() obey EG(precision), I'd say we should retain that behavior in
these explicit conversions. That way existing (functional) uses
aren't impacted, we just quietly start working on other cases (where
scientific notation and/or non-period delimiters would have been
used).
Before proceeding to an RFC (if necessary at all), I'm looking forward
to hear your thoughts regarding this topic.
IMO, this falls under fixing an existing bug (bc*() functions don't
accept certain types of floats), so it wouldn't need an RFC to be
accepted. I would wait a little bit to get feedback on the scale
question before merging though. I would even offer that we could
apply this to 7.[012] safely.
-Sara
I already have proposed to introduce bcinit() which would take a string,
int or float, and return a properly formatted string with the given or
default scale, see https://github.com/php/php-src/pull/2746. Nikita
suggested to accept the arguments allowed bybcinit()for all
functions in the first place. This would, however, raise the question
which scale should be applied to floats.I agree with Nikita. Given that current behavior for floats passed to
bc*() obey EG(precision), I'd say we should retain that behavior in
these explicit conversions. That way existing (functional) uses
aren't impacted, we just quietly start working on other cases (where
scientific notation and/or non-period delimiters would have been
used).
Okay, that makes sense.
Before proceeding to an RFC (if necessary at all), I'm looking forward
to hear your thoughts regarding this topic.IMO, this falls under fixing an existing bug (bc*() functions don't
accept certain types of floats), so it wouldn't need an RFC to be
accepted. I would wait a little bit to get feedback on the scale
question before merging though. I would even offer that we could
apply this to 7.[012] safely.
Hmm, the bc*() functions are documented to accept strings, and as such
there is no actual bug; see also Mike's final assessment on
https://bugs.php.net/55160. If we would consider this as bug, we also
had to consider some other cases regarding our peculiar float to string
casting behavior as bug. And even if we would, we still couldn't
generally change the float to string cast for BC reasons easily (if ever).
Anyhow, I'll gladly wait for further comments. :)
--
Christoph M. Becker