v1.35.0 breaks any program trying to use a `-v` flag

[marcind]

The following simple program breaks when dd-trace-go is upgraded to v1.35.0

package main

import (
	"flag"
	"os"

	"gopkg.in/DataDog/dd-trace-go.v1/ddtrace/tracer"
)

func main() {
	var v bool
	flag.BoolVar(&v, "v", false, "make verbose")
	flag.Parse()

	if v {
		println("Hello world")
	}

	if os.Getenv("DD_AGENT_HOST") != "" {
		tracer.Start()
		defer tracer.Stop()
	}
}

The output when executing go run main.go is something like

/var/folders/yg/7bk34drs0zjfsb9b_r_1sykr0000gn/T/go-build2940344100/b001/exe/main flag redefined: v
panic: /var/folders/yg/7bk34drs0zjfsb9b_r_1sykr0000gn/T/go-build2940344100/b001/exe/main flag redefined: v

goroutine 1 [running]:
flag.(*FlagSet).Var(0xc000078180, {0x14e0a78, 0xc000026658}, {0x1450a20, 0x1}, {0x145389a, 0xc})
        /go/src/flag/flag.go:879 +0x2f4
flag.BoolVar(...)
        /go/src/flag/flag.go:638
main.main()
        /repro/main.go:12 +0x70
exit status 2

This has been traced to the usage of the github.com/golang/glog package (which also uses the -v flag)

Glog is imported via the following chain

go mod why github.com/golang/glog
# github.com/golang/glog
example.com/repro
gopkg.in/DataDog/dd-trace-go.v1/ddtrace/tracer
github.com/DataDog/datadog-agent/pkg/obfuscate
github.com/dgraph-io/ristretto
github.com/dgraph-io/ristretto/z
github.com/golang/glog

The breaking change was introduced in #1069 (committed in https://github.com/DataDog/dd-trace-go/commit/f55a62275c129771aec581fa9263f89259bf2f16).

Rolling back to the previous commit by executing go get gopkg.in/DataDog/dd-trace-go.v1/ddtrace/tracer@11c835cb0 && go mod tidy makes the above-mentioned program work again.

It is not my expectation that some transitive dependency of a transitive dependency would cause applications to break because some library thinks it knows better and decides to hijack a bunch of flags (including something as generic as -v).

[marcind]

For what it's worth it appears that Google has no intention of fixing the underlying problem and libraries should just avoid depending on the glog package.

[felixge]

@marcind sorry you're hitting this issue. I think it's reasonable to expect that upgrading dd-trace-go doesn't break your application.

That being said, it's a tricky situation. glog seems difficult to prune from our dependency graph without giving up on ristretto. We also can't use a replace directive to point to a fork from within dd-trace-go since that only works for the main module according to the module spec. @marcind do you see any elegant workarounds for us?

On your end it seems like you should be able to use a FlagSet rather than relying on pkg-global state as a workaround?

cc @gbbr who worked on the bits related to glog

[marcind]

@felixge seems like the easiest might be to lobby for https://github.com/dgraph-io/ristretto/pull/292 and/or https://github.com/dgraph-io/ristretto/pull/293 as it appears dd-trace-go is not the only project affected by glog.

Outside of that, have you considered not using ristretto?

[radykal-com]

This can now be fixed by updating github.com/DataDog/datadog-agent/pkg/obfuscate version. As per this PR https://github.com/DataDog/datadog-agent/pull/15005 they replaced the ristretto dependency with a maintained fork.

Any chance to fix it?

[felixge]

@radykal-com the problem should be fixed in main now, and will be released soon (ETA end of month, maybe early next month).

Sorry this took so long and for the problems it has caused.