bin/find-can-be-private-symbols.py

   1 #!/usr/bin/python3
   2 #
   3 # Find exported symbols that can be made non-exported.
   4 #
   5 # Noting that (a) parsing these commands is a pain, the output is quite irregular and (b) I'm fumbling in the
   6 # dark here, trying to guess what exactly constitutes an "import" vs an "export" of a symbol, linux linking
   7 # is rather complex.
   8 #
   9 # Takes about 5min to run on a decent machine.
  10 #
  11 # The standalone function analysis is reasonable reliable, but the class/method analysis is less so
  12 #   (something to do with destructor thunks not showing up in my results?)
  13 #
  14 # Also, the class/method analysis will not catch problems like
  15 #    'dynamic_cast from 'Foo' with hidden type visibility to 'Bar' with default type visibility'
  16 #    but loplugin:dyncastvisibility will do that for you
  17 #
  18
  19 import subprocess
  20 import re
  21
  22 exported_symbols1 = set()
  23 imported_symbols1 = set()
  24 exported_symbols2 = set() # decoded
  25 imported_symbols2 = set() # decoded
  26
  27
  28 # find all our shared libs
  29 subprocess_find = subprocess.Popen("find ./instdir -name *.so && find ./workdir/LinkTarget/CppunitTest -name *.so",
  30     stdout=subprocess.PIPE, shell=True)
  31 with subprocess_find.stdout as txt:
  32     for line in txt:
  33         sharedlib = line.strip()
  34         # look for exported symbols
  35         subprocess_nm = subprocess.Popen(b"nm -D " + sharedlib, stdout=subprocess.PIPE, shell=True)
  36         with subprocess_nm.stdout as txt2:
  37             # We are looking for lines something like:
  38             # 0000000000036ed0 T flash_component_getFactory
  39             line_regex = re.compile(r'^[0-9a-fA-F]+ T ')
  40             for line2_bytes in txt2:
  41                 line2 = line2_bytes.strip().decode("utf-8")
  42                 if line_regex.match(line2):
  43                     sym = line2.split(" ")[2].strip()
  44                     exported_symbols1.add(sym)
  45         subprocess_nm.terminate()
  46         # look for imported symbols
  47         subprocess_objdump = subprocess.Popen(b"objdump -T " + sharedlib, stdout=subprocess.PIPE, shell=True)
  48         with subprocess_objdump.stdout as txt2:
  49             # ignore some header bumpf
  50             txt2.readline()
  51             txt2.readline()
  52             txt2.readline()
  53             txt2.readline()
  54             # We are looking for lines something like:
  55             # 0000000000000000      DF *UND*  0000000000000000     _ZN16FilterConfigItem10WriteInt32ERKN3rtl8OUStringEi
  56             for line2_bytes in txt2:
  57                 line2 = line2_bytes.strip().decode("utf-8")
  58                 if "*UND*" not in line2: continue
  59                 tokens = line2.split(" ")
  60                 sym = tokens[len(tokens)-1].strip()
  61                 imported_symbols1.add(sym)
  62         subprocess_objdump.terminate()
  63 subprocess_find.terminate()
  64
  65 # look for imported symbols in executables
  66 subprocess_find = subprocess.Popen("find ./instdir -name *.bin", stdout=subprocess.PIPE, shell=True)
  67 with subprocess_find.stdout as txt:
  68     for line in txt:
  69         executable = line.strip()
  70         # look for exported symbols
  71         subprocess_nm = subprocess.Popen(b"nm -D " + executable + b" | grep -w U", stdout=subprocess.PIPE, shell=True)
  72         with subprocess_nm.stdout as txt2:
  73             # We are looking for lines something like:
  74             # U sal_detail_deinitialize
  75             for line2_bytes in txt2:
  76                 line2 = line2_bytes.strip().decode("utf-8")
  77                 sym = line2.split(" ")[1]
  78                 imported_symbols1.add(sym)
  79 subprocess_find.terminate()
  80
  81 # Now we have to symbolize before comparing because sometimes (due to thunks) two
  82 # different encoded names symbolize to the same method/func name
  83 #
  84 progress = 0
  85 progress_max_len = len(imported_symbols1) + len(exported_symbols1)
  86 for sym in imported_symbols1:
  87     progress += 1
  88     if (progress % 128 == 0): print( str(int(progress * 100 / progress_max_len)) + "%")
  89     filtered_sym = subprocess.check_output(["c++filt", sym]).strip().decode("utf-8")
  90     if filtered_sym.startswith("non-virtual thunk to "): filtered_sym = filtered_sym[21:]
  91     elif filtered_sym.startswith("virtual thunk to "): filtered_sym = filtered_sym[17:]
  92     imported_symbols2.add(filtered_sym)
  93 progress = 0
  94 for sym in exported_symbols1:
  95     progress += 1
  96     if (progress % 128 == 0): print( str(int(progress * 100 / progress_max_len)) + "%")
  97     filtered_sym = subprocess.check_output(["c++filt", sym]).strip().decode("utf-8")
  98     if filtered_sym.startswith("non-virtual thunk to "): filtered_sym = filtered_sym[21:]
  99     elif filtered_sym.startswith("virtual thunk to "): filtered_sym = filtered_sym[17:]
 100     exported_symbols2.add(filtered_sym)
 101
 102 unused_exports = exported_symbols2 - imported_symbols2
 103 print("exported       = " + str(len(exported_symbols2)))
 104 print("imported       = " + str(len(imported_symbols2)))
 105 print("unused_exports = " + str(len(unused_exports)))
 106
 107 # for each class, count how many symbols will become hidden if we mark the class as hidden
 108 can_be_hidden_count = dict()
 109 for sym in exported_symbols2:
 110     i = sym.rfind("::")
 111     if i == -1: continue
 112     clz = sym[:i]
 113     if clz in can_be_hidden_count:
 114         can_be_hidden_count[clz] = can_be_hidden_count[clz] + 1
 115     else:
 116         can_be_hidden_count[clz] = 1
 117 for sym in imported_symbols2:
 118     i = sym.rfind("::")
 119     if i == -1: continue
 120     clz = sym[:i]
 121     if clz in can_be_hidden_count:
 122         can_be_hidden_count[clz] = can_be_hidden_count[clz] - 1
 123     else:
 124         can_be_hidden_count[clz] = -1
 125 # convert to list, and sort the results in descending order
 126 can_be_hidden_list = list()
 127 for clz in can_be_hidden_count:
 128     cnt = can_be_hidden_count[clz]
 129     if cnt > 0:
 130         can_be_hidden_list.append((cnt, clz))
 131 can_be_hidden_list.sort(reverse=True)
 132 with open("bin/find-can-be-private-symbols.classes.results", "wt") as f:
 133     for i in can_be_hidden_list:
 134         if i[0] < 10: break
 135         f.write(str(i[0]) + " " + i[1] + "\n")
 136
 137
 138 with open("bin/find-can-be-private-symbols.functions.results", "wt") as f:
 139     for sym in sorted(unused_exports):
 140         # Filter out most of the noise.
 141         # No idea where these are coming from, but not our code.
 142         if sym.startswith("CERT_"): continue
 143         elif sym.startswith("DER_"): continue
 144         elif sym.startswith("FORM_"): continue
 145         elif sym.startswith("FPDF"): continue
 146         elif sym.startswith("HASH_"): continue
 147         elif sym.startswith("Hunspell_"): continue
 148         elif sym.startswith("LL_"): continue
 149         elif sym.startswith("LP_"): continue
 150         elif sym.startswith("LU"): continue
 151         elif sym.startswith("MIP"): continue
 152         elif sym.startswith("MPS"): continue
 153         elif sym.startswith("NSS"): continue
 154         elif sym.startswith("NSC_"): continue
 155         elif sym.startswith("PK11"): continue
 156         elif sym.startswith("PL_"): continue
 157         elif sym.startswith("PQ"): continue
 158         elif sym.startswith("PBE_"): continue
 159         elif sym.startswith("PORT_"): continue
 160         elif sym.startswith("PRP_"): continue
 161         elif sym.startswith("PR_"): continue
 162         elif sym.startswith("PT_"): continue
 163         elif sym.startswith("QS_"): continue
 164         elif sym.startswith("REPORT_"): continue
 165         elif sym.startswith("RSA_"): continue
 166         elif sym.startswith("SEC"): continue
 167         elif sym.startswith("SGN"): continue
 168         elif sym.startswith("SOS"): continue
 169         elif sym.startswith("SSL_"): continue
 170         elif sym.startswith("VFY_"): continue
 171         elif sym.startswith("_PR_"): continue
 172         elif sym.startswith("ber_"): continue
 173         elif sym.startswith("bfp_"): continue
 174         elif sym.startswith("ldap_"): continue
 175         elif sym.startswith("ne_"): continue
 176         elif sym.startswith("opj_"): continue
 177         elif sym.startswith("pg_"): continue
 178         elif sym.startswith("pq"): continue
 179         elif sym.startswith("presolve_"): continue
 180         elif sym.startswith("sqlite3_"): continue
 181         elif sym.startswith("libepubgen::"): continue
 182         elif sym.startswith("lucene::"): continue
 183         elif sym.startswith("Hunspell::"): continue
 184         elif sym.startswith("sk_"): continue
 185         elif sym.startswith("_Z"): continue
 186         # dynamically loaded
 187         elif sym.endswith("get_implementation"): continue
 188         elif sym.endswith("component_getFactory"): continue
 189         elif sym == "CreateUnoWrapper": continue
 190         elif sym == "ExportDOC": continue
 191         elif sym == "ExportRTF": continue
 192         elif sym == "GetSaveWarningOfMSVBAStorage_ww8": continue
 193         elif sym == "GetSpecialCharsForEdit": continue
 194         elif sym.startswith("Import"): continue
 195         elif sym.startswith("Java_com_sun_star_"): continue
 196         elif sym.startswith("TestImport"): continue
 197         elif sym.startswith("getAllCalendars_"): continue
 198         elif sym.startswith("getAllCurrencies_"): continue
 199         elif sym.startswith("getAllFormats"): continue
 200         elif sym.startswith("getBreakIteratorRules_"): continue
 201         elif sym.startswith("getCollationOptions_"): continue
 202         elif sym.startswith("getCollatorImplementation_"): continue
 203         elif sym.startswith("getContinuousNumberingLevels_"): continue
 204         elif sym.startswith("getDateAcceptancePatterns_"): continue
 205         elif sym.startswith("getForbiddenCharacters_"): continue
 206         elif sym.startswith("getIndexAlgorithm_"): continue
 207         elif sym.startswith("getLCInfo_"): continue
 208         elif sym.startswith("getLocaleItem_"): continue
 209         elif sym.startswith("getOutlineNumberingLevels_"): continue
 210         elif sym.startswith("getReservedWords_"): continue
 211         elif sym.startswith("getSTC_"): continue
 212         elif sym.startswith("getSearchOptions_"): continue
 213         elif sym.startswith("getTransliterations_"): continue
 214         elif sym.startswith("getUnicodeScripts_"): continue
 215         elif sym.startswith("lok_"): continue
 216         # UDK API
 217         elif sym.startswith("osl_"): continue
 218         elif sym.startswith("rtl_"): continue
 219         elif sym.startswith("typelib_"): continue
 220         elif sym.startswith("typereg_"): continue
 221         elif sym.startswith("uno_"): continue
 222         # remove things we found that do not exist in our source code, they're not ours
 223         #if not(extractFunctionNameFromSignature(sym) in all_source_names): continue
 224         f.write(sym + "\n")
 225