Ever wondered what's making your ELF or Mach-O binary big? Bloaty McBloatface will show you a size profile of the binary so you can understand what's taking up space inside.
Bloaty works on binaries, shared objects, object files, and
static libraries (.a
files). It supports ELF/DWARF and
Mach-O, though the Mach-O support is much more preliminary
(it shells out to otool
/symbols
instead of parsing the
file directly).
This is not an official Google product.
To build, simply run:
$ make
Bloaty depends on RE2, so the Makefile will download it (via a Git submodule) and build that also.
To run the tests (requires that cmake
is installed and
available on your path) run:
$ make test
Run it directly on a binary target. For example, run it on itself.
$ ./bloaty bloaty
On Linux you'll see output something like:
VM SIZE FILE SIZE
-------------- --------------
0.0% 0 .debug_info 2.97Mi 38.3%
0.0% 0 .debug_loc 2.30Mi 29.7%
0.0% 0 .debug_str 1.03Mi 13.3%
0.0% 0 .debug_ranges 611Ki 7.7%
72.8% 332Ki .text 332Ki 4.2%
0.0% 0 .debug_line 218Ki 2.8%
0.0% 0 .debug_abbrev 85.4Ki 1.1%
0.0% 0 .strtab 62.8Ki 0.8%
13.2% 60.0Ki .rodata 60.0Ki 0.8%
7.0% 31.8Ki .eh_frame 31.8Ki 0.4%
0.0% 0 .symtab 27.8Ki 0.3%
0.0% 0 .debug_aranges 13.5Ki 0.2%
2.3% 10.5Ki .gcc_except_table 10.5Ki 0.1%
1.5% 6.77Ki [Other] 5.60Ki 0.1%
0.9% 4.18Ki .eh_frame_hdr 4.18Ki 0.1%
0.8% 3.54Ki .dynsym 3.54Ki 0.0%
0.8% 3.52Ki .dynstr 3.52Ki 0.0%
0.7% 2.98Ki .rela.plt 2.98Ki 0.0%
0.1% 568 [ELF Headers] 2.93Ki 0.0%
0.0% 34 [Unmapped] 2.85Ki 0.0%
0.0% 4 [None] 0 0.0%
100.0% 456Ki TOTAL 7.75Mi 100.0%
The "VM SIZE" column tells you how much space the binary will take when it is loaded into memory. The "FILE SIZE" column tells you about how much space the binary is taking on disk. These two can be very different from each other:
- Some data lives in the file but isn't loaded into memory, like debug information.
- Some data is mapped into memory but doesn't exist in the
file. This mainly applies to the
.bss
section (zero-initialized data).
The default breakdown in Bloaty is by sections, but many other ways of slicing the binary are supported such as symbols and segments. If you compiled with debug info, you can even break down by compile units and inlines!
$ ./bloaty bloaty -d compileunits
VM SIZE FILE SIZE
-------------- --------------
27.9% 128Ki [None] 7.43Mi 95.9%
12.9% 59.2Ki src/bloaty.cc 59.0Ki 0.7%
7.3% 33.4Ki re2/re2.cc 32.3Ki 0.4%
6.9% 31.6Ki re2/dfa.cc 31.6Ki 0.4%
6.8% 31.4Ki re2/parse.cc 31.4Ki 0.4%
6.7% 30.9Ki src/dwarf.cc 30.9Ki 0.4%
6.7% 30.6Ki re2/regexp.cc 27.8Ki 0.4%
5.1% 23.7Ki re2/compile.cc 23.7Ki 0.3%
4.3% 19.7Ki re2/simplify.cc 19.7Ki 0.2%
3.2% 14.8Ki src/elf.cc 14.8Ki 0.2%
3.1% 14.2Ki re2/nfa.cc 14.2Ki 0.2%
1.8% 8.34Ki re2/bitstate.cc 8.34Ki 0.1%
1.7% 7.84Ki re2/prog.cc 7.84Ki 0.1%
1.6% 7.13Ki re2/tostring.cc 7.13Ki 0.1%
1.5% 6.67Ki re2/onepass.cc 6.67Ki 0.1%
1.4% 6.58Ki src/macho.cc 6.58Ki 0.1%
0.7% 3.27Ki src/main.cc 3.27Ki 0.0%
0.2% 797 [Other] 797 0.0%
0.1% 666 util/stringprintf.cc 666 0.0%
0.1% 573 util/strutil.cc 573 0.0%
0.1% 476 util/rune.cc 476 0.0%
100.0% 460Ki TOTAL 7.75Mi 100.0%
Run Bloaty with --help
to see a list of available options:
$ ./bloaty --help
Bloaty McBloatface: a size profiler for binaries.
USAGE: bloaty [options] file... [-- base_file...]
Options:
-d <sources> Comma-separated list of sources to scan.
-n <num> How many rows to show per level before collapsing
other keys into '[Other]'. Set to '0' for unlimited.
Defaults to 20.
-r <regex> Add regex to the list of regexes.
Format for regex is:
SOURCE:s/PATTERN/REPLACEMENT/
-s <sortby> Whether to sort by VM or File size. Possible values
are:
-s vm
-s file
-s both (the default: sorts by max(vm, file)).
-v Verbose output. Dumps warnings encountered during
processing and full VM/file maps at the end.
Add more v's (-vv, -vvv) for even more.
--help Display this message and exit.
--list-sources Show a list of available sources and exit.
You can use Bloaty to see how the size of a binary changed.
On the command-line, pass --
followed by the files you
want to use as the diff base.
For example, here is a size diff between a couple different versions of Bloaty, showing how it grew when I added some features.
$ ./bloaty bloaty -- oldbloaty
VM SIZE FILE SIZE
++++++++++++++ GROWING ++++++++++++++
[ = ] 0 .debug_str +41.2Ki +5.0%
[ = ] 0 .debug_info +36.8Ki +1.3%
[ = ] 0 .debug_loc +12.4Ki +0.6%
+1.8% +6.12Ki .text +6.12Ki +1.8%
[ = ] 0 .debug_ranges +4.47Ki +0.8%
[ = ] 0 .debug_line +2.69Ki +1.3%
[ = ] 0 .strtab +1.52Ki +3.1%
+3.9% +1.32Ki .eh_frame +1.32Ki +3.9%
+1.6% +1.12Ki .rodata +1.12Ki +1.6%
[ = ] 0 .symtab +696 +2.3%
[ = ] 0 .debug_aranges +288 +2.4%
+2.7% +272 .gcc_except_table +272 +2.7%
+2.7% +136 .eh_frame_hdr +136 +2.7%
+1.2% +48 .dynsym +48 +1.2%
+1.4% +48 .rela.plt +48 +1.4%
+1.4% +32 .plt +32 +1.4%
+0.6% +22 .dynstr +22 +0.6%
+1.3% +16 .got.plt +16 +1.3%
+1.2% +4 .gnu.version +4 +1.2%
-------------- SHRINKING --------------
-18.5% -10 [Unmapped] -1.14Ki -31.4%
[ = ] 0 .debug_abbrev -72 -0.1%
+1.9% +9.12Ki TOTAL +107Ki +1.5%
Each line shows the how much each part changed compared to its previous size. The "TOTAL" line shows how much the size changed overall.
Bloaty supports breaking the binary down in lots of
different ways. You can combine multiple data sources into
a single hierarchical profile. For example, we can use the
segments
and sections
data sources in a single report:
$ bloaty -d segments,sections bloaty
VM SIZE FILE SIZE
-------------- --------------
0.0% 0 [Unmapped] 7.31Mi 94.2%
-NAN% 0 .debug_info 2.97Mi 40.6%
-NAN% 0 .debug_loc 2.30Mi 31.5%
-NAN% 0 .debug_str 1.03Mi 14.2%
-NAN% 0 .debug_ranges 611Ki 8.2%
-NAN% 0 .debug_line 218Ki 2.9%
-NAN% 0 .debug_abbrev 85.4Ki 1.1%
-NAN% 0 .strtab 62.8Ki 0.8%
-NAN% 0 .symtab 27.8Ki 0.4%
-NAN% 0 .debug_aranges 13.5Ki 0.2%
-NAN% 0 [Unmapped] 2.82Ki 0.0%
-NAN% 0 .shstrtab 371 0.0%
-NAN% 0 .comment 43 0.0%
99.2% 452Ki LOAD [RX] 452Ki 5.7%
73.4% 332Ki .text 332Ki 73.4%
13.3% 60.0Ki .rodata 60.0Ki 13.3%
7.0% 31.8Ki .eh_frame 31.8Ki 7.0%
2.3% 10.5Ki .gcc_except_table 10.5Ki 2.3%
0.9% 4.18Ki .eh_frame_hdr 4.18Ki 0.9%
0.8% 3.54Ki .dynsym 3.54Ki 0.8%
0.8% 3.52Ki .dynstr 3.52Ki 0.8%
0.7% 2.98Ki .rela.plt 2.98Ki 0.7%
0.4% 2.00Ki .plt 2.00Ki 0.4%
0.1% 568 [ELF Headers] 568 0.1%
0.1% 408 .rela.dyn 408 0.1%
0.1% 304 .gnu.version_r 304 0.1%
0.1% 302 .gnu.version 302 0.1%
0.0% 216 .gnu.hash 216 0.0%
0.0% 36 .note.gnu.build-id 36 0.0%
0.0% 32 .note.ABI-tag 32 0.0%
0.0% 28 .interp 28 0.0%
0.0% 26 .init 26 0.0%
0.0% 18 [Unmapped] 18 0.0%
0.0% 9 .fini 9 0.0%
0.8% 3.46Ki LOAD [RW] 1.88Ki 0.0%
45.6% 1.58Ki .bss 0 0.0%
29.3% 1.02Ki .got.plt 1.02Ki 54.1%
14.9% 528 .dynamic 528 27.4%
7.1% 252 .data 252 13.1%
1.4% 48 .init_array 48 2.5%
0.7% 24 .got 24 1.2%
0.5% 16 [Unmapped] 16 0.8%
0.2% 8 .fini_array 8 0.4%
0.2% 8 .jcr 8 0.4%
0.1% 4 [None] 0 0.0%
0.0% 0 [ELF Headers] 2.38Ki 0.0%
100.0% 456Ki TOTAL 7.75Mi 100.0%
Bloaty displays a maximum of 20 lines for each level; other
values are grouped into an [Other]
bin. Use -n <num>
to override this setting. If you pass -n 0
, all data
will be output without collapsing anything into [Other]
.
Bloaty has many data sources built in. It's easy to add a new data source if you have a new way of mapping address ranges to some interesting higher-level abstraction.
While Bloaty works on binaries, shared objects, object
files, and static libraries (.a
files), some of the data
sources don't work on object files. This applies especially
to data sources that read debug info.
Segments are what the run-time loader uses to determine what
parts of the binary need to be loaded/mapped into memory.
There are usually just a few segments: one for each set of
mmap()
permissions required:
$ bloaty -d segments bloaty
VM SIZE FILE SIZE
-------------- --------------
0.0% 0 [Unmapped] 7.31Mi 94.2%
99.2% 452Ki LOAD [RX] 452Ki 5.7%
0.8% 3.46Ki LOAD [RW] 1.88Ki 0.0%
0.0% 0 [ELF Headers] 2.38Ki 0.0%
100.0% 456Ki TOTAL 7.75Mi 100.0%
Here we see one segment mapped [R E]
(read/execute) and
one segment mapped [RW ]
(read/write). A large part of
the binary is not loaded into memory, which we see as
[Unmapped]
.
Object files and static libraries don't have segments. However we fake it by grouping sections by their flags. This gives us a break-down sort of like real segments.
$ ./bloaty bloaty -d segments src/bloaty.o
VM SIZE FILE SIZE
-------------- --------------
0.0% 0 [Unmapped] 7.31Mi 67.6%
0.0% 0 Section [] 2.95Mi 27.3%
85.2% 452Ki LOAD [RX] 452Ki 4.1%
11.3% 59.8Ki Section [AX] 59.8Ki 0.5%
0.0% 0 [ELF Headers] 28.3Ki 0.3%
2.9% 15.4Ki Section [A] 15.4Ki 0.1%
0.7% 3.46Ki LOAD [RW] 1.88Ki 0.0%
0.0% 41 Section [AW] 20 0.0%
100.0% 531Ki TOTAL 10.8Mi 100.0%
Sections give us a bit more granular look into the binary. If we want to find the symbol table, the unwind information, or the debug information, each kind of information lives in its own section. Bloaty's default output is sections.
$ bloaty -d sections bloaty
VM SIZE FILE SIZE
-------------- --------------
0.0% 0 .debug_info 2.97Mi 38.3%
0.0% 0 .debug_loc 2.30Mi 29.7%
0.0% 0 .debug_str 1.03Mi 13.3%
0.0% 0 .debug_ranges 611Ki 7.7%
72.8% 332Ki .text 332Ki 4.2%
0.0% 0 .debug_line 218Ki 2.8%
0.0% 0 .debug_abbrev 85.4Ki 1.1%
0.0% 0 .strtab 62.8Ki 0.8%
13.2% 60.0Ki .rodata 60.0Ki 0.8%
7.0% 31.8Ki .eh_frame 31.8Ki 0.4%
0.0% 0 .symtab 27.8Ki 0.3%
0.0% 0 .debug_aranges 13.5Ki 0.2%
2.3% 10.5Ki .gcc_except_table 10.5Ki 0.1%
1.5% 6.77Ki [Other] 5.60Ki 0.1%
0.9% 4.18Ki .eh_frame_hdr 4.18Ki 0.1%
0.8% 3.54Ki .dynsym 3.54Ki 0.0%
0.8% 3.52Ki .dynstr 3.52Ki 0.0%
0.7% 2.98Ki .rela.plt 2.98Ki 0.0%
0.1% 568 [ELF Headers] 2.93Ki 0.0%
0.0% 34 [Unmapped] 2.85Ki 0.0%
0.0% 4 [None] 0 0.0%
100.0% 456Ki TOTAL 7.75Mi 100.0%
Symbols come from the symbol table, and represent individual functions or variables. C++ symbols are demangled for convenience.
$ ./bloaty -d symbols bloaty
VM SIZE FILE SIZE
-------------- --------------
17.9% 81.9Ki [Unmapped] 7.39Mi 95.3%
62.3% 283Ki [Other] 284Ki 3.6%
2.7% 12.3Ki re2::RE2::Match(re2::StringPiece const&, int, int, re2::RE2::Anchor, re2::String 12.3Ki 0.2%
1.7% 7.83Ki re2::unicode_groups 7.83Ki 0.1%
1.7% 7.56Ki re2::NFA::Search 7.56Ki 0.1%
1.3% 5.76Ki re2::BitState::TrySearch 5.76Ki 0.1%
1.2% 5.43Ki bloaty::Bloaty::ScanAndRollupFile 5.43Ki 0.1%
1.0% 4.49Ki re2::DFA::DFA 4.49Ki 0.1%
1.0% 4.35Ki bool bloaty::(anonymous namespace)::ForEachElf<bloaty::(anonymous namespace)::Do 4.35Ki 0.1%
1.0% 4.34Ki re2::Regexp::Parse 4.34Ki 0.1%
0.9% 4.20Ki re2::RE2::Init 4.20Ki 0.1%
0.9% 4.09Ki re2::Prog::IsOnePass 4.09Ki 0.1%
0.9% 4.04Ki re2::Compiler::PostVisit 4.04Ki 0.1%
0.9% 4.04Ki bloaty::ReadDWARFInlines 4.04Ki 0.1%
0.9% 3.91Ki re2::Regexp::FactorAlternationRecursive 3.91Ki 0.0%
0.8% 3.77Ki re2::DFA::RunStateOnByte 3.77Ki 0.0%
0.8% 3.68Ki re2::unicode_casefold 3.68Ki 0.0%
0.8% 3.52Ki bloaty::ElfFileHandler::ProcessFile 3.52Ki 0.0%
0.7% 3.40Ki re2::DFA::InlinedSearchLoop(re2::DFA::SearchParams*, bool, bool, bool) [clone .c 3.40Ki 0.0%
0.7% 3.38Ki re2::DFA::InlinedSearchLoop(re2::DFA::SearchParams*, bool, bool, bool) [clone .c 3.38Ki 0.0%
0.0% 165 [None] 0 0.0%
100.0% 456Ki TOTAL 7.75Mi 100.0%
When you pass multiple files to Bloaty, the inputfiles
source will let you break it down by input file:
$ ./bloaty -d inputfiles src/*.o
VM SIZE FILE SIZE
-------------- --------------
51.8% 75.2Ki src/bloaty.o 3.05Mi 48.2%
28.2% 40.9Ki src/dwarf.o 2.04Mi 32.2%
12.1% 17.5Ki src/elf.o 579Ki 8.9%
5.5% 7.99Ki src/macho.o 415Ki 6.4%
2.5% 3.57Ki src/main.o 279Ki 4.3%
100.0% 145Ki TOTAL 6.34Mi 100.0%
When you are running Bloaty on a .a
file, the armembers
source will let you break it down by .o
file inside the
archive.
./bloaty -d armembers src/libbloaty.a
VM SIZE FILE SIZE
-------------- --------------
53.1% 75.2Ki bloaty.o 3.05Mi 50.1%
28.9% 40.9Ki dwarf.o 2.04Mi 33.5%
12.4% 17.5Ki elf.o 579Ki 9.3%
5.6% 7.99Ki macho.o 415Ki 6.7%
0.0% 0 [AR Symbol Table] 27.3Ki 0.4%
0.0% 0 [AR Headers] 308 0.0%
100.0% 141Ki TOTAL 6.10Mi 100.0%
You are free to use this data source even for non-.a
files, but it won't be very useful since it will always just
resolve to the input file (the .a
file).
Using debug information, we can tell what compile unit (and
corresponding source file) each bit of the binary came from.
There are a couple different places in DWARF we can look for
this information; currently we mainly use the
.debug_aranges
section. It's not perfect and sometimes
you'll see some of the binary show up as [None]
if it's
not mentioned in aranges (improving this is a TODO). But it
can tell us a lot.
$ ./bloaty -d compileunits bloaty
VM SIZE FILE SIZE
-------------- --------------
27.9% 128Ki [None] 7.43Mi 95.9%
12.9% 59.2Ki src/bloaty.cc 59.0Ki 0.7%
7.3% 33.4Ki re2/re2.cc 32.3Ki 0.4%
6.9% 31.6Ki re2/dfa.cc 31.6Ki 0.4%
6.8% 31.4Ki re2/parse.cc 31.4Ki 0.4%
6.7% 30.9Ki src/dwarf.cc 30.9Ki 0.4%
6.7% 30.6Ki re2/regexp.cc 27.8Ki 0.4%
5.1% 23.7Ki re2/compile.cc 23.7Ki 0.3%
4.3% 19.7Ki re2/simplify.cc 19.7Ki 0.2%
3.2% 14.8Ki src/elf.cc 14.8Ki 0.2%
3.1% 14.2Ki re2/nfa.cc 14.2Ki 0.2%
1.8% 8.34Ki re2/bitstate.cc 8.34Ki 0.1%
1.7% 7.84Ki re2/prog.cc 7.84Ki 0.1%
1.6% 7.13Ki re2/tostring.cc 7.13Ki 0.1%
1.5% 6.67Ki re2/onepass.cc 6.67Ki 0.1%
1.4% 6.58Ki src/macho.cc 6.58Ki 0.1%
0.7% 3.27Ki src/main.cc 3.27Ki 0.0%
0.2% 797 [Other] 797 0.0%
0.1% 666 util/stringprintf.cc 666 0.0%
0.1% 573 util/strutil.cc 573 0.0%
0.1% 476 util/rune.cc 476 0.0%
100.0% 460Ki TOTAL 7.75Mi 100.0%
The DWARF debugging information also contains "line info" information that understands inlining. So within a function, it will know which instructions came from an inlined function from a header file. This is the information the debugger uses to point at a specific source line as you're tracing through a program.
$ ./bloaty -d inlines bloaty
VM SIZE FILE SIZE
-------------- --------------
2.4% 110Ki [None] 7.42Mi 95.6%
90.3% 4.01Mi /usr/include/c++/4.8/bitsstl_vector.h:414 15.3Ki 0.2%
5.5% 250Ki [Other] 250Ki 3.2%
0.3% 11.4Ki /usr/include/c++/4.8/bitsbasic_string.h:539 11.4Ki 0.1%
0.2% 8.81Ki /usr/include/c++/4.8ostream:535 8.81Ki 0.1%
0.2% 7.59Ki /usr/include/c++/4.8/bitsbasic_ios.h:456 7.59Ki 0.1%
0.1% 6.20Ki /usr/include/c++/4.8streambuf:466 6.20Ki 0.1%
0.1% 6.06Ki /usr/include/c++/4.8/bitsbasic_string.h:249 6.06Ki 0.1%
0.1% 4.24Ki /usr/include/c++/4.8/bitsbasic_string.h:240 4.24Ki 0.1%
0.1% 3.61Ki /usr/include/c++/4.8/bitsbasic_ios.h:276 3.61Ki 0.0%
0.1% 3.51Ki /usr/include/c++/4.8/extatomicity.h:81 3.51Ki 0.0%
0.1% 3.19Ki /usr/include/c++/4.8/bitsbasic_string.h:583 3.19Ki 0.0%
0.1% 3.06Ki /usr/include/c++/4.8/bitsbasic_string.h:293 3.06Ki 0.0%
0.1% 2.94Ki /usr/include/c++/4.8/extnew_allocator.h:110 2.94Ki 0.0%
0.1% 2.89Ki /usr/include/c++/4.8ostream:385 2.89Ki 0.0%
0.1% 2.87Ki /usr/include/c++/4.8/bitsstl_construct.h:102 2.87Ki 0.0%
0.1% 2.86Ki /usr/include/c++/4.8/extatomicity.h:84 2.86Ki 0.0%
0.1% 2.76Ki /usr/include/c++/4.8/extatomicity.h:49 2.76Ki 0.0%
0.1% 2.70Ki /usr/include/c++/4.8/bitschar_traits.h:271 2.70Ki 0.0%
0.1% 2.62Ki /usr/include/c++/4.8/bitsbasic_string.h:275 2.62Ki 0.0%
0.1% 2.58Ki /usr/include/c++/4.8ostream:93 2.58Ki 0.0%
100.0% 4.45Mi TOTAL 7.75Mi 100.0%
You can filter the lists by using regular expressions. For example, to view by source file by group all re2-related sources together, you can write:
(TODO: this appears to be broken at the moment, needs fixing!)
$ ./bloaty -d inlines bloaty -r 'inlines:s/.*re2.*/RE2/'
VM SIZE FILE SIZE
-------------- --------------
42.5% 200k [None] 6.36M 96.0%
44.7% 211k RE2 211k 3.1%
8.5% 40.1k bloaty.cc 40.0k 0.6%
2.3% 10.6k elf.cc 10.6k 0.2%
1.5% 7.16k dwarf.cc 7.16k 0.1%
0.1% 666 util/stringprintf.cc 666 0.0%
0.1% 590 macho.cc 590 0.0%
0.1% 573 util/strutil.cc 573 0.0%
0.1% 476 util/rune.cc 476 0.0%
0.1% 287 util/hash.cc 287 0.0%
0.0% 96 util/valgrind.cc 96 0.0%
100.0% 472k TOTAL 6.63M 100.0%
Note: this functionality is a bit under-developed and subject to change. For example, there is not yet a way to escape backslashes.
Here are some tentative plans for future features.
If we can analyze references between symbols, this would enable a lot of features:
- Detect garbage symbols (ie. how much would the binary
shrink if we compiled with
-ffunction-sections -fdata-sections -Wl,-gc-sections
). - Understand why a particular symbol can't be
garbage-collected (like
ld -why_live
on OS X). - Visualize the dependency tree of symbols (probably as a dominator tree) so users can see the weight of their binary in this way.
One of the things we have to do in Bloaty is deal with
incomplete information. For examples, .debug_aranges
which we use for the compileunits
data source is often
missing or incomplete. Refining the input sources to be
more complete and accurate will make help Bloaty's numbers
be even more accurate.