[AIX] Limit `extract_symbols.py`'s Pools Size to 8

[qiongsiwu1]

Summary

On AIX, extract_symbols.py's default pool size may be too big for the logical partition(LPAR)'s allocated memory, and may lead to intermittent IO errors. This patch limits the pool size to 8 to avoid such errors.

[qiongsiwu1]

Ping for review. Thanks!

[qiongsiwu1]

Ping for review. Thanks so much!

[thieta]

Looks fine to me, I have no experience with AIX but I don't see a problem with this fix.

[qiongsiwu1]

In D157884#4602822, @thieta wrote:

Looks fine to me, I have no experience with AIX but I don't see a problem with this fix.

Thanks for the review @thieta !

llvm/utils/extract_symbols.py

 
 from __future__ import print_function
 import sys
 import re
 import os
 import subprocess
 import multiprocessing
 import argparse
+import platform
 
 # Define a function which extracts a list of pairs of (symbols, is_def) from a
 # library using llvm-nm becuase it can work both with regular and bitcode files.
 # We use subprocess.Popen and yield a symbol at a time instead of using
 # subprocess.check_output and returning a list as, especially on Windows, waiting
 # for the entire output to be ready can take a significant amount of time.
 def nm_get_symbols(tool, lib):
     # '-P' means the output is in portable format,
if __name__ == "__main__":
 
     # Check if calling convention decoration is used by inspecting the first
     # library in the list
     calling_convention_decoration = readobj_is_32bit_windows(args.readobj, libs[0])
 
     # Extract symbols from libraries in parallel. This is a huge time saver when
     # doing a debug build, as there are hundreds of thousands of symbols in each
     # library.
-    pool = multiprocessing.Pool()
+    # FIXME: On AIX, the default pool size can be too big for a logical
+    #        partition's allocated memory, and can lead to an out of memory
+    #        IO error. We are setting the pool size to 8 to avoid such
+    #        errors at the moment, and will look for a graceful solution later.
+    pool = multiprocessing.Pool(8) if platform.system() == "AIX" \
+                                   else multiprocessing.Pool()
     try:
         # Only one argument can be passed to the mapping function, and we can't
         # use a lambda or local function definition as that doesn't work on
         # windows, so create a list of tuples which duplicates the arguments
         # that are the same in all calls.
         vals = [
             (args.nm, should_keep_symbol, calling_convention_decoration, x)
             for x in libs