[BUG] Slow compile times for string arguments in device functions (0.22 vs 0.24)

[Jlisowskyy]

Bug description
When using strings as device function arguments, compilation time explodes

Steps to reproduce bug

1. Prepare a clean python environment
2. Execute pip install numba-cuda
3. The code below demonstrates the behaviour:

import numpy as np
from numba import cuda
import sys

@cuda.jit(device=True, forceinline=True)
def load_slow(gmem, value):
    if value == "1":
        gmem[cuda.threadIdx.x] = 1.0
    else:
        gmem[cuda.threadIdx.x] = 0.0

@cuda.jit(device=True, forceinline=True)
def load(gmem, value):
    if value == 1.0:
        gmem[cuda.threadIdx.x] = 1.0
    else:
        gmem[cuda.threadIdx.x] = 0.0

def get_kernel(use_slow):
    if use_slow:
        @cuda.jit
        def slow(buff):
            load_slow(buff, "1")
            
        return slow
        
    @cuda.jit
    def quick(buff):
        load(buff, 1.0)
        
    return quick

def main():
    arg = sys.argv[1] if len(sys.argv) > 1 else "slow"
    kernel = get_kernel(arg == "slow")
    
    threads = 32

    buff = np.zeros((threads,), dtype=np.float32)
    buff_d = cuda.to_device(buff)

    kernel[1, threads](buff_d)
    cuda.synchronize()
    
    buff_result = buff_d.copy_to_host()
    
    assert np.linalg.norm(buff_result - np.ones((threads,), dtype=np.float32)) == 0.0
    print("finished")

if __name__ == "__main__":
    main()

Expected behavior
Using string arguments slows down compilation significantly (~2s vs ~2min):

time python .test.dir.c/slow.py quick

real    0m2.536s
user    0m2.442s
sys     0m0.166s

time python .test.dir.c/slow.py slow

real    1m57.233s
user    1m31.162s
sys     0m26.162s

Environment details:

• Environment location: wsl local machine
• Method of numba-cuda install: pip install
• Python env:

numba-cuda==0.24.0
numba==0.63.1
Python 3.11.14

Additional context
This issue does not appear in numba-cuda==0.22. It appears that a redirect (in the error module), added in numba-cuda==0.22.1 triggers this behaviour.

Command log:

$ python3.11 -m venv .slow
$ source .slow/bin/activate
$ pip cache purge
$ rm -rf ~/.nv/ComputeCache/
$ find . -type d -name __pycache__ -exec rm -rf {} +
$ find . -name "*.pyc" -delete
$ rm -rf ~/.cache/numba
$ pip install numba-cuda
$ time python .test.dir.c/slow.py quick

real    0m2.536s
user    0m2.442s
sys     0m0.166s
$ time python .test.dir.c/slow.py slow

real    1m57.233s
user    1m31.162s
sys     0m26.162s
$ pip install numba-cuda==0.22
$ time python .test.dir.c/slow.py slow

real    0m3.242s
user    0m3.095s
sys     0m0.274s