libm
When a math function detects an error, it sets the static variable
errno. Depending upon the severity of the error, it may also
print a message to stderr. None of this behavior is reentrant.
When one process detects an error, it sets errno. If another
process is testing errno, it detects the change and probably
fails. Note that failing system calls can also set errno. This
problem can only be fixed by either ignoring errno, or treating
it as part of the context of a process and switching it along with the
rest of a processor state. In normal debugged programs, there are
usually no math subroutine errors--and therefore no matherr
calls; in that situation, the math functions behave reentrantly.
As an alternative, you can use the reentrant versions of the mathematical functions: these versions have a different name, normally formed by prepending `_' and appending `_r', and use an extra argument--a pointer to the particular reentrancy structure to use. See section `Reentrancy' in The Cygnus C Support Library, for more discussion of this approach to reentrancy.
The reentrancy structure is always an additional first argument; for example, the reentrant version of `double acos (double x)' is `double _acos_r (void *reent, double x)'.
Here is a list of the names for reentrant versions of the mathematical library functions:
_acos_r _gammaf_r_r _log10f_r _acosf_r _hypot_r _log_r _acosh_r _hypotf_r _logf_r _acoshf_r _j0_r _pow_r _asin_r _j0f_r _powf_r _asinf_r _j1_r _remainder_r _atanh_r _j1f_r _sinh_r _atanhf_r _jn_r _sinhf_r _cosh_r _jnf_r _sqrt_r _coshf_r _ldexp_r _sqrtf_r _exp_r _ldexpf_r _y0_r _expf_r _lgamma_r _y0f_r _fmod_r _lgamma_r_r _y1_r _fmodf_r _lgammaf_r _y1f_r _gamma_r _lgammaf_r_r _yn_r _gamma_r_r _log10_r _ynf_r _gammaf_r