-d compileunits on go binary shows suspiciously large size for sync/atomic

[psanford]

I ran bloaty on an internal go binary which reported sync/atomic using ~30% of the binary size. This seemed... unlikely. So I tried running bloatly on an etcd binary (linux amd64) compiled from master which shows something very similar:

$ /tmp/bloaty/bloaty -d compileunits etcd -n 5
     VM SIZE                              FILE SIZE
 --------------                        --------------
  49.0%  2.92Mi [142 Others]            4.27Mi  46.2%
  30.8%  1.83Mi sync/atomic             2.79Mi  30.2%
   9.1%   557Ki net/http                 824Ki   8.7%
   7.3%   444Ki runtime                  733Ki   7.8%
   3.8%   234Ki net                      345Ki   3.7%
   0.0%       0 [section .debug_frame]   329Ki   3.5%
 100.0%  5.96Mi TOTAL                   9.23Mi 100.0%

30% doesn't sound right to me as sync/atomic is a small amount of assembly code and not much else.

I'm hoping someone could shed some light onto what is going on here. Is this somehow related to inlining?

[haberman]

I'm not a Go user. How can I reproduce this?

[psanford]

The etcd binary I was examining is here: https://s3.amazonaws.com/www.sanford.io/etcd

To build the binary yourself (assuming you have the go tool installed):

git clone https://github.com/coreos/etcd
cd etcd
./build
ls bin/

[dominikh]

Possibly related to #176

[haberman]

Sorry I was not able to debug this earlier.

I just tried your command and I am getting reasonable results (I did have to specify non-compressed dwarf per #204):

$ ../bloaty -d compileunits bin/etcd -n 30
    FILE SIZE        VM SIZE    
 --------------  -------------- 
  36.3%  16.3Mi  31.9%  6.62Mi    [318 Others]
  12.3%  5.52Mi  26.6%  5.52Mi    [section .gopclntab]
   9.3%  4.17Mi   4.0%   848Ki    runtime
   4.9%  2.19Mi   3.7%   790Ki    go.etcd.io/etcd/api/v3/etcdserverpb
   3.5%  1.56Mi   3.3%   701Ki    net/http
   2.9%  1.31Mi   2.3%   483Ki    github.com/gogo/protobuf/proto
   2.5%  1.12Mi   4.8%  1.00Mi    vendor/golang.org/x/text/unicode/norm
   2.4%  1.09Mi   2.4%   511Ki    go.etcd.io/etcd/server/v3/etcdserver
   2.4%  1.08Mi   0.0%       0    [section .debug_frame]
   2.2%   988Ki   1.9%   401Ki    crypto/tls
   1.9%   881Ki   1.6%   345Ki    github.com/json-iterator/go
   1.8%   845Ki   1.5%   315Ki    github.com/golang/protobuf/proto
   1.5%   667Ki   1.3%   275Ki    net
   1.2%   569Ki   1.2%   248Ki    google.golang.org/grpc
   1.1%   505Ki   0.9%   196Ki    go.etcd.io/etcd/client/v3
   1.1%   500Ki   0.9%   194Ki    google.golang.org/grpc/internal/transport
   1.1%   489Ki   0.9%   183Ki    math/big
   1.0%   458Ki   0.6%   136Ki    github.com/spf13/pflag
   1.0%   438Ki   0.7%   156Ki    golang.org/x/net/http2
   0.9%   412Ki   0.8%   167Ki    github.com/gogo/protobuf/protoc-gen-gogo/descriptor
   0.9%   407Ki   0.9%   188Ki    go.etcd.io/etcd/server/v3/etcdserver/api/rafthttp
   0.9%   407Ki   0.9%   190Ki    go.etcd.io/etcd/api/v3/etcdserverpb/gw
   0.9%   394Ki   0.6%   133Ki    go.etcd.io/etcd/server/v3/etcdmain
   0.9%   392Ki   0.8%   170Ki    gopkg.in/yaml.v2
   0.8%   383Ki   0.7%   142Ki    reflect
   0.8%   349Ki   0.6%   123Ki    github.com/modern-go/reflect2
   0.8%   348Ki   0.4%  90.0Ki    github.com/prometheus/client_golang/prometheus/promhttp
   0.8%   348Ki   1.6%   348Ki    [section .rodata]
   0.7%   323Ki   0.6%   128Ki    internal/profile
   0.7%   317Ki   0.7%   149Ki    go.uber.org/zap/zapcore
   0.7%   309Ki   0.6%   137Ki    crypto/x509
 100.0%  44.8Mi 100.0%  20.7Mi    TOTAL

I'll close this now. If you have a case where Bloaty is still returning unreasonable results, please let me know.