bcpow issue with large exponent

[devnexen]

Description

The following code:

<?php
bcpow(10, PHP_INT_MAX, 1);
?>

Resulted in this output:

<After a dozen of seconds>
PHP Fatal error:  Allowed memory size of 18446744073709551615 bytes exhausted

But I expected this output instead:

PHP Version

PHP >=8.2

Operating System

No response

[nielsdos]

Isn't this expected? You'd need log10(10**PHP_INT_MAX) = PHP_INT_MAX digits, which is that same amount in bytes.

[nielsdos]

Oh I get it, you mean because the final result only has scale 1...

[nielsdos]

Note that the scale only influences the number of digits after the comma.

[devnexen]

Note that I know next to nothing about bcmath. If the underlying library can't handle such exponent (note that my simple script froze for seconds) it might be best to throw exception. But again, I do not know better, feel free to close if you think it is irrelevant.

[nielsdos]

The reason it hangs is because it's doing multiplications during that time, and once it comes to the point that we need more digits than we have memory it bails.
It's possible to compute upfront how much memory we need and avoid doing the computation if we're gonna bail, however I'm not sure it makes sense to optimize the "sad flow". cc @Girgias @SakiTakamachi

[jorgsowa]

If the underlying library

At this point there is no underlying library, but the implementation of the mathematical operations. It's been heavily improved, and some operations completely differ from the initial implementation.

I fixed a similar issue with bcpow() last year: #11761 In your example we know upfront what's the result of the calculation. But then what cases we should treat exclusively, 10 as a base? What about 100, or 0.1. For me, it would be difficult to set a boundary.

[devnexen]

GMP which I know a little bit better chose the "fail fast" route instead. e.g. in the same case, the underlying libgmp raises explicitly a SIGFPE (can be reproduced with PHP < 8.2.26), in other cases when the code consumer is being "irreasonable" asking libgmp large allocations, it aborts. Not necessarily great neither ... but now PHP catches arguments and throws exceptions, I was trying to see if we can make the bcmath extension consistent to some degree.

[nielsdos]

We can detect it upfront, the number of bytes required is equal to the number of digits in the result. The number of digits is log10(base**exponent) which is equal to exponent*log10(base); or if you want it as an integer: floor(exponent*log10(base))+1

[SakiTakamachi]

To display this result, a huge amount of bytes is required regardless of the scale value, so I think the current specifications are fine.

[nielsdos]

I'd agree. I'm gonna close it then. Thanks.