Bug report

Bug description:

>>>importmath>>>math.ldexp(6993274598585239,-1126)5e-324>>>

The correct result would be 1e-323. This is obviously a bug in the Windows ldexp implementation (it works fine on Linux). But it would be good, if it could be fixed on the CPython side with a workaround.
math.ldexp is used by mpmath to round from multiprecision to float, leading to incorrect rounding on Windows.

CPython versions tested on:

3.13

Operating systems tested on:

Windows

Linked PRs

• gh-132876: workaround broken ldexp() on Windows 10 #133135
• [3.14] gh-132876: workaround broken ldexp() on Windows 10 (GH-133135) #134684
• [3.13] gh-132876: workaround broken ldexp() on Windows 10 (GH-133135) #134685