strconv: `ParseFloat` silently assumes bitSize=64 for any value that is not 32

[colega]

What version of Go are you using (go version)?

$ go version 1.17.3

Does this issue reproduce with the latest release?

Yes

What operating system and processor architecture are you using (go env)?

go env Output

$ go env
GO111MODULE=""
GOARCH="amd64"
GOBIN="/home/oleg/go/bin"
GOCACHE="/home/oleg/.cache/go-build"
GOENV="/home/oleg/.config/go/env"
GOEXE=""
GOEXPERIMENT=""
GOFLAGS=""
GOHOSTARCH="amd64"
GOHOSTOS="linux"
GOINSECURE=""
GOMODCACHE="/home/oleg/go/pkg/mod"
GONOPROXY="github.com/grafana"
GONOSUMDB="github.com/grafana"
GOOS="linux"
GOPATH="/home/oleg/go"
GOPRIVATE="github.com/grafana"
GOPROXY="https://proxy.golang.org,direct"
GOROOT="/usr/local/go"
GOSUMDB="sum.golang.org"
GOTMPDIR=""
GOTOOLDIR="/usr/local/go/pkg/tool/linux_amd64"
GOVCS=""
GOVERSION="go1.17.3"
GCCGO="gccgo"
AR="ar"
CC="gcc"
CXX="g++"
CGO_ENABLED="1"
GOMOD="/home/oleg/w/github.com/grafana/mimir/go.mod"
CGO_CFLAGS="-g -O2"
CGO_CPPFLAGS=""
CGO_CXXFLAGS="-g -O2"
CGO_FFLAGS="-g -O2"
CGO_LDFLAGS="-g -O2"
PKG_CONFIG="pkg-config"
GOGCCFLAGS="-fPIC -m64 -pthread -fmessage-length=0 -fdebug-prefix-map=/tmp/go-build1481112582=/tmp/go-build -gno-record-gcc-switches"

What did you do?

I called strconv.ParseFloat with bitSize=10. It succeeded.

What did you expect to see?

I expected it to fail, as the documentation says it only accepts bitSize=32 and bitSize=64.

What did you see instead?

It didn't fail, but parsed my number.

Why is this an issue

strconv.ParseFloat silently assumes bitSize=64 for any bitSize value that is not 32. This means that changing a call from strconv.ParseFloat("1.1", 16) to strconv.ParseFloat("1.1", 32) actually decreases the precision instead of increasing it.

In my particular case, I made a similar change (suggested by a linter, because honestly, I didn't know about bitSize limitation), and it resulted in a flaky test because of the precision decrease.

Further thoughts

I am not brave enough to open a PR returning an error when strconv.ParseFloat() is called with bitSize values that are not 32/64, as I feel that woud break the internet at this point, however I do think that the function should either comply with it's comment, fail, or at least have this behavior documented.

[mknyszek]

CC @rsc @griesemer @ianlancetaylor via https://dev.golang.org/owners

[ianlancetaylor]

We tried to fix this in https://go.dev/cl/248219 for #40706. However, that led to #42297, which showed that enforcing the documented rules caused too many problems with existing code. We considered a vet check in #42389, but decided against it.

We could update the documentation, but I think that in general we should prefer that people pass the correct bit size, even though other sizes happen to work.

So, we aren't going to do anything here. Sorry. Thanks for reporting the problem.