office-gobmx/bin/find-can-be-private-symbols.py
Noel Grandin e10e5c15cc improve the script for reducing symbol exports
And apply some of the results

Change-Id: If555476fdd951cbc1d01fb3ef3ab1cbca2b64960
Reviewed-on: https://gerrit.libreoffice.org/c/core/+/124896
Tested-by: Jenkins
Reviewed-by: Noel Grandin <noel.grandin@collabora.co.uk>
2021-12-01 12:13:59 +01:00

222 lines
10 KiB
Python
Executable file

#!/usr/bin/python3
#
# Find exported symbols that can be made non-exported.
#
# Noting that (a) parsing these commands is a pain, the output is quite irregular and (b) I'm fumbling in the
# dark here, trying to guess what exactly constitutes an "import" vs an "export" of a symbol, linux linking
# is rather complex.
#
# Takes about 5min to run on a decent machine.
#
# The standalone function analysis is reasonable reliable, but the class/method analysis is less so
# (something to do with destructor thunks not showing up in my results?)
#
# Also, the class/method analysis will not catch problems like
# 'dynamic_cast from 'Foo' with hidden type visibility to 'Bar' with default type visibility'
# but loplugin:dyncastvisibility will do that for you
#
import subprocess
import sys
import re
exported_symbols1 = set()
imported_symbols1 = set()
exported_symbols2 = set() # decoded
imported_symbols2 = set() # decoded
# all names that exist in the source code
#all_source_names = set()
#subprocess_find_all_source_names = subprocess.Popen("git grep -oh -P '\\b\\w\\w\\w+\\b' -- '*.h*' | sort -u",
# stdout=subprocess.PIPE, stderr=subprocess.STDOUT, shell=True)
#with subprocess_find_all_source_names.stdout as txt:
# for line in txt:
# line = line.strip()
# all_source_names.add(line)
#subprocess_find_all_source_names.terminate()
# find all our shared libs
subprocess_find = subprocess.Popen("find ./instdir -name *.so && find ./workdir/LinkTarget/CppunitTest -name *.so",
stdout=subprocess.PIPE, shell=True)
with subprocess_find.stdout as txt:
for line in txt:
sharedlib = line.strip()
# look for exported symbols
subprocess_nm = subprocess.Popen(b"nm -D " + sharedlib, stdout=subprocess.PIPE, shell=True)
with subprocess_nm.stdout as txt2:
# We are looking for lines something like:
# 0000000000036ed0 T flash_component_getFactory
line_regex = re.compile(r'^[0-9a-fA-F]+ T ')
for line2_bytes in txt2:
line2 = line2_bytes.strip().decode("utf-8")
if line_regex.match(line2):
sym = line2.split(" ")[2].strip()
exported_symbols1.add(sym)
subprocess_nm.terminate()
# look for imported symbols
subprocess_objdump = subprocess.Popen(b"objdump -T " + sharedlib, stdout=subprocess.PIPE, shell=True)
with subprocess_objdump.stdout as txt2:
# ignore some header bumpf
txt2.readline()
txt2.readline()
txt2.readline()
txt2.readline()
# We are looking for lines something like:
# 0000000000000000 DF *UND* 0000000000000000 _ZN16FilterConfigItem10WriteInt32ERKN3rtl8OUStringEi
for line2_bytes in txt2:
line2 = line2_bytes.strip().decode("utf-8")
if not("*UND*"in line2): continue
tokens = line2.split(" ")
sym = tokens[len(tokens)-1].strip()
imported_symbols1.add(sym)
subprocess_objdump.terminate()
subprocess_find.terminate()
# look for imported symbols in executables
subprocess_find = subprocess.Popen("find ./instdir -name *.bin", stdout=subprocess.PIPE, shell=True)
with subprocess_find.stdout as txt:
for line in txt:
executable = line.strip()
# look for exported symbols
subprocess_nm = subprocess.Popen(b"nm -D " + executable + b" | grep -w U", stdout=subprocess.PIPE, shell=True)
with subprocess_nm.stdout as txt2:
# We are looking for lines something like:
# U sal_detail_deinitialize
for line2_bytes in txt2:
line2 = line2_bytes.strip().decode("utf-8")
sym = line2.split(" ")[1]
imported_symbols1.add(sym)
subprocess_find.terminate()
#progress = 0;
#for sym in sorted(imported_symbols - exported_symbols):
# progress += 1
# if (progress % 128 == 0): print( str(int(progress * 100 / len(diff))) + "%")
# filtered_sym = subprocess.check_output(["c++filt", sym]).strip().decode("utf-8")
# if filtered_sym.startswith("non-virtual thunk to "): filtered_sym = filtered_sym[21:]
# elif filtered_sym.startswith("virtual thunk to "): filtered_sym = filtered_sym[17:]
# print("Symbol imported but not exported? " + filtered_sym)
# Now we have to symbolize before comparing because sometimes (due to thunks) two
# different encoded names symbolize to the same method/func name
#
progress = 0;
progress_max_len = len(imported_symbols1) + len(exported_symbols1)
for sym in imported_symbols1:
progress += 1
if (progress % 128 == 0): print( str(int(progress * 100 / progress_max_len)) + "%")
filtered_sym = subprocess.check_output(["c++filt", sym]).strip().decode("utf-8")
if filtered_sym.startswith("non-virtual thunk to "): filtered_sym = filtered_sym[21:]
elif filtered_sym.startswith("virtual thunk to "): filtered_sym = filtered_sym[17:]
imported_symbols2.add(filtered_sym)
progress = 0;
for sym in exported_symbols1:
progress += 1
if (progress % 128 == 0): print( str(int(progress * 100 / progress_max_len)) + "%")
filtered_sym = subprocess.check_output(["c++filt", sym]).strip().decode("utf-8")
if filtered_sym.startswith("non-virtual thunk to "): filtered_sym = filtered_sym[21:]
elif filtered_sym.startswith("virtual thunk to "): filtered_sym = filtered_sym[17:]
exported_symbols2.add(filtered_sym)
unused_exports = exported_symbols2 - imported_symbols2
print("exported = " + str(len(exported_symbols2)))
print("imported = " + str(len(imported_symbols2)))
print("unused_exports = " + str(len(unused_exports)))
#def extractFunctionNameFromSignature(sym):
# i = sym.find("(")
# if i == -1: return sym
# return sym[:i]
with open("bin/find-can-be-private-symbols.functions.results", "wt") as f:
for sym in sorted(unused_exports):
# Filter out most of the noise.
# No idea where these are coming from, but not our code.
if sym.startswith("CERT_"): continue
elif sym.startswith("DER_"): continue
elif sym.startswith("FORM_"): continue
elif sym.startswith("FPDF"): continue
elif sym.startswith("HASH_"): continue
elif sym.startswith("Hunspell_"): continue
elif sym.startswith("LL_"): continue
elif sym.startswith("LP_"): continue
elif sym.startswith("LU"): continue
elif sym.startswith("MIP"): continue
elif sym.startswith("MPS"): continue
elif sym.startswith("NSS"): continue
elif sym.startswith("NSC_"): continue
elif sym.startswith("PK11"): continue
elif sym.startswith("PL_"): continue
elif sym.startswith("PQ"): continue
elif sym.startswith("PBE_"): continue
elif sym.startswith("PORT_"): continue
elif sym.startswith("PRP_"): continue
elif sym.startswith("PR_"): continue
elif sym.startswith("PT_"): continue
elif sym.startswith("QS_"): continue
elif sym.startswith("REPORT_"): continue
elif sym.startswith("RSA_"): continue
elif sym.startswith("SEC"): continue
elif sym.startswith("SGN"): continue
elif sym.startswith("SOS"): continue
elif sym.startswith("SSL_"): continue
elif sym.startswith("VFY_"): continue
elif sym.startswith("_PR_"): continue
elif sym.startswith("ber_"): continue
elif sym.startswith("bfp_"): continue
elif sym.startswith("ldap_"): continue
elif sym.startswith("ne_"): continue
elif sym.startswith("opj_"): continue
elif sym.startswith("pg_"): continue
elif sym.startswith("pq"): continue
elif sym.startswith("presolve_"): continue
elif sym.startswith("sqlite3_"): continue
elif sym.startswith("libepubgen::"): continue
elif sym.startswith("lucene::"): continue
elif sym.startswith("Hunspell::"): continue
elif sym.startswith("sk_"): continue
elif sym.startswith("_Z"): continue
# dynamically loaded
elif sym.endswith("get_implementation"): continue
elif sym.endswith("component_getFactory"): continue
elif sym == "CreateDialogFactory": continue
elif sym == "CreateUnoWrapper": continue
elif sym == "CreateWindow": continue
elif sym == "ExportDOC": continue
elif sym == "ExportPPT": continue
elif sym == "ExportRTF": continue
elif sym == "GetSaveWarningOfMSVBAStorage_ww8": continue
elif sym == "GetSpecialCharsForEdit": continue
elif sym.startswith("Import"): continue
elif sym.startswith("Java_com_sun_star_"): continue
elif sym.startswith("TestImport"): continue
elif sym.startswith("getAllCalendars_"): continue
elif sym.startswith("getAllCurrencies_"): continue
elif sym.startswith("getAllFormats"): continue
elif sym.startswith("getBreakIteratorRules_"): continue
elif sym.startswith("getCollationOptions_"): continue
elif sym.startswith("getCollatorImplementation_"): continue
elif sym.startswith("getContinuousNumberingLevels_"): continue
elif sym.startswith("getDateAcceptancePatterns_"): continue
elif sym.startswith("getForbiddenCharacters_"): continue
elif sym.startswith("getIndexAlgorithm_"): continue
elif sym.startswith("getLCInfo_"): continue
elif sym.startswith("getLocaleItem_"): continue
elif sym.startswith("getOutlineNumberingLevels_"): continue
elif sym.startswith("getReservedWords_"): continue
elif sym.startswith("getSTC_"): continue
elif sym.startswith("getSearchOptions_"): continue
elif sym.startswith("getTransliterations_"): continue
elif sym.startswith("getUnicodeScripts_"): continue
elif sym.startswith("lok_"): continue
# UDK API
elif sym.startswith("osl_"): continue
elif sym.startswith("rtl_"): continue
elif sym.startswith("typelib_"): continue
elif sym.startswith("typereg_"): continue
elif sym.startswith("uno_"): continue
# remove things we found that do not exist in our source code, they're not ours
#if not(extractFunctionNameFromSignature(sym) in all_source_names): continue
f.write(sym + "\n")