1
0
mirror of https://github.com/golang/go synced 2024-11-22 20:04:47 -07:00
go/doc/godebug.md

353 lines
16 KiB
Markdown
Raw Normal View History

---
title: "Go, Backwards Compatibility, and GODEBUG"
layout: article
---
<!--
This document is kept in the Go repo, not x/website,
because it documents the full list of known GODEBUG settings,
which are tied to a specific release.
-->
## Introduction {#intro}
Go's emphasis on backwards compatibility is one of its key strengths.
There are, however, times when we cannot maintain complete compatibility.
If code depends on buggy (including insecure) behavior,
then fixing the bug will break that code.
New features can also have similar impacts:
enabling the HTTP/2 use by the HTTP client broke programs
connecting to servers with buggy HTTP/2 implementations.
These kinds of changes are unavoidable and
[permitted by the Go 1 compatibility rules](/doc/go1compat).
Even so, Go provides a mechanism called GODEBUG to
reduce the impact such changes have on Go developers
using newer toolchains to compile old code.
A GODEBUG setting is a `key=value` pair
that controls the execution of certain parts of a Go program.
The environment variable `GODEBUG`
can hold a comma-separated list of these settings.
For example, if a Go program is running in an environment that contains
GODEBUG=http2client=0,http2server=0
then that Go program will disable the use of HTTP/2 by default in both
the HTTP client and the HTTP server.
It is also possible to set the default `GODEBUG` for a given program
(discussed below).
When preparing any change that is permitted by Go 1 compatibility
but may nonetheless break some existing programs,
we first engineer the change to keep as many existing programs working as possible.
For the remaining programs,
we define a new GODEBUG setting that
allows individual programs to opt back in to the old behavior.
A GODEBUG setting may not be added if doing so is infeasible,
but that should be extremely rare.
GODEBUG settings added for compatibility will be maintained
for a minimum of two years (four Go releases).
Some, such as `http2client` and `http2server`,
will be maintained much longer, even indefinitely.
When possible, each GODEBUG setting has an associated
[runtime/metrics](/pkg/runtime/metrics/) counter
named `/godebug/non-default-behavior/<name>:events`
that counts the number of times a particular program's
behavior has changed based on a non-default value
for that setting.
For example, when `GODEBUG=http2client=0` is set,
`/godebug/non-default-behavior/http2client:events`
counts the number of HTTP transports that the program
has configured without HTTP/2 support.
## Default GODEBUG Values {#default}
When a GODEBUG setting is not listed in the environment variable,
its value is derived from three sources:
the defaults for the Go toolchain used to build the program,
amended to match the Go version listed in `go.mod`,
and then overridden by explicit `//go:debug` lines in the program.
The [GODEBUG History](#history) gives the exact defaults for each Go toolchain version.
For example, Go 1.21 introduces the `panicnil` setting,
controlling whether `panic(nil)` is allowed;
it defaults to `panicnil=0`, making `panic(nil)` a run-time error.
Using `panicnil=1` restores the behavior of Go 1.20 and earlier.
When compiling a work module or workspace that declares
an older Go version, the Go toolchain amends its defaults
to match that older Go version as closely as possible.
For example, when a Go 1.21 toolchain compiles a program,
if the work module's `go.mod` or the workspace's `go.work`
says `go` `1.20`, then the program defaults to `panicnil=1`,
matching Go 1.20 instead of Go 1.21.
Because this method of setting GODEBUG defaults was introduced only in Go 1.21,
programs listing versions of Go earlier than Go 1.20 are configured to match Go 1.20,
not the older version.
To override these defaults, starting in Go 1.23, the work module's `go.mod`
or the workspace's `go.work` can list one or more `godebug` lines:
godebug (
default=go1.21
panicnil=1
asynctimerchan=0
)
The special key `default` indicates a Go version to take unspecified
settings from. This allows setting the GODEBUG defaults separately
from the Go language version in the module.
In this example, the program is asking for Go 1.21 semantics and
then asking for the old pre-Go 1.21 `panic(nil)` behavior and the
new Go 1.23 `asynctimerchan=0` behavior.
Only the work module's `go.mod` is consulted for `godebug` directives.
Any directives in required dependency modules are ignored.
It is an error to list a `godebug` with an unrecognized setting.
(Toolchains older than Go 1.23 reject all `godebug` lines, since they do not
understand `godebug` at all.)
The defaults from the `go` and `godebug` lines apply to all main
packages that are built. For more fine-grained control,
starting in Go 1.21, a main package's source files
can include one or more `//go:debug` directives at the top of the file
(preceding the `package` statement).
The `godebug` lines in the previous example would be written:
//go:debug default=go1.21
//go:debug panicnil=1
//go:debug asynctimerchan=0
Starting in Go 1.21, the Go toolchain treats a `//go:debug` directive
with an unrecognized GODEBUG setting as an invalid program.
Programs with more than one `//go:debug` line for a given setting
are also treated as invalid.
(Older toolchains ignore `//go:debug` directives entirely.)
The defaults that will be compiled into a main package
are reported by the command:
{{raw `
go list -f '{{.DefaultGODEBUG}}' my/main/package
`}}
Only differences from the base Go toolchain defaults are reported.
When testing a package, `//go:debug` lines in the `*_test.go`
files are treated as directives for the test's main package.
In any other context, `//go:debug` lines are ignored by the toolchain;
`go` `vet` reports such lines as misplaced.
## GODEBUG History {#history}
This section documents the GODEBUG settings introduced and removed in each major Go release
for compatibility reasons.
Packages or programs may define additional settings for internal debugging purposes;
for example,
see the [runtime documentation](/pkg/runtime#hdr-Environment_Variables)
and the [go command documentation](/cmd/go#hdr-Build_and_test_caching).
### Go 1.23
time: avoid stale receives after Timer/Ticker Stop/Reset return A proposal discussion in mid-2020 on #37196 decided to change time.Timer and time.Ticker so that their Stop and Reset methods guarantee that no old value (corresponding to the previous configuration of the Timer or Ticker) will be received after the method returns. The trivial way to do this is to make the Timer/Ticker channels unbuffered, create a goroutine per Timer/Ticker feeding the channel, and then coordinate with that goroutine during Stop/Reset. Since Stop/Reset coordinate with the goroutine and the channel is unbuffered, there is no possibility of a stale value being sent after Stop/Reset returns. Of course, we do not want an extra goroutine per Timer/Ticker, but that's still a good semantic model: behave like the channels are unbuffered and fed by a coordinating goroutine. The actual implementation is more effort but behaves like the model. Specifically, the timer channel has a 1-element buffer like it always has, but len(t.C) and cap(t.C) are special-cased to return 0 anyway, so user code cannot see what's in the buffer except with a receive. Stop/Reset lock out any stale sends and then clear any pending send from the buffer. Some programs will change behavior. For example: package main import "time" func main() { t := time.NewTimer(2 * time.Second) time.Sleep(3 * time.Second) if t.Reset(2*time.Second) != false { panic("expected timer to have fired") } <-t.C <-t.C } This program (from #11513) sleeps 3s after setting a 2s timer, resets the timer, and expects Reset to return false: the Reset is too late and the send has already occurred. It then expects to receive two values: the one from before the Reset, and the one from after the Reset. With an unbuffered timer channel, it should be clear that no value can be sent during the time.Sleep, so the time.Reset returns true, indicating that the Reset stopped the timer from going off. Then there is only one value to receive from t.C: the one from after the Reset. In 2015, I used the above example as an argument against this change. Note that a correct version of the program would be: func main() { t := time.NewTimer(2 * time.Second) time.Sleep(3 * time.Second) if !t.Reset(2*time.Second) { <-t.C } <-t.C } This works with either semantics, by heeding t.Reset's result. The change should not affect correct programs. However, one way that the change would be visible is when programs use len(t.C) (instead of a non-blocking receive) to poll whether the timer has triggered already. We might legitimately worry about breaking such programs. In 2020, discussing #37196, Bryan Mills and I surveyed programs using len on timer channels. These are exceedingly rare to start with; nearly all the uses are buggy; and all the buggy programs would be fixed by the new semantics. The details are at [1]. To further reduce the impact of this change, this CL adds a temporary GODEBUG setting, which we didn't know about yet in 2015 and 2020. Specifically, asynctimerchan=1 disables the change and is the default for main programs in modules that use a Go version before 1.23. We hope to be able to retire this setting after the minimum 2-year window. Setting asynctimerchan=1 also disables the garbage collection change from CL 568341, although users shouldn't need to know that since it is not a semantically visible change (unless we have bugs!). As an undocumented bonus that we do not officially support, asynctimerchan=2 disables the channel buffer change but keeps the garbage collection change. This may help while we are shaking out bugs in either of them. Fixes #37196. [1] https://github.com/golang/go/issues/37196#issuecomment-641698749 Change-Id: I8925d3fb2b86b2ae87fd2acd055011cbf7bd5916 Reviewed-on: https://go-review.googlesource.com/c/go/+/568341 Reviewed-by: Austin Clements <austin@google.com> Auto-Submit: Russ Cox <rsc@golang.org> LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
2024-02-29 20:39:49 -07:00
Go 1.23 changed the channels created by package time to be unbuffered
(synchronous), which makes correct use of the [`Timer.Stop`](/pkg/time/#Timer.Stop)
and [`Timer.Reset`](/pkg/time/#Timer.Reset) method results much easier.
The [`asynctimerchan` setting](/pkg/time/#NewTimer) disables this change.
There are no runtime metrics for this change,
This setting may be removed in a future release, Go 1.27 at the earliest.
time: garbage collect unstopped Tickers and Timers From the beginning of Go, the time package has had a gotcha: if you use a select on <-time.After(1*time.Minute), even if the select finishes immediately because some other case is ready, the underlying timer from time.After keeps running until the minute is over. This pins the timer in the timer heap, which keeps it from being garbage collected and in extreme cases also slows down timer operations. The lack of garbage collection is the more important problem. The docs for After warn against this scenario and suggest using NewTimer with a call to Stop after the select instead, purely to work around this garbage collection problem. Oddly, the docs for NewTimer and NewTicker do not mention this problem, but they have the same issue: they cannot be collected until either they are Stopped or, in the case of Timer, the timer expires. (Tickers repeat, so they never expire.) People have built up a shared knowledge that timers and tickers need to defer t.Stop even though the docs do not mention this (it is somewhat implied by the After docs). This CL fixes the garbage collection problem, so that a timer that is unreferenced can be GC'ed immediately, even if it is still running. The approach is to only insert the timer into the heap when some channel operation is blocked on it; the last channel operation to stop using the timer takes it back out of the heap. When a timer's channel is no longer referenced, there are no channel operations blocked on it, so it's not in the heap, so it can be GC'ed immediately. This CL adds an undocumented GODEBUG asynctimerchan=1 that will disable the change. The documentation happens in the CL 568341. Fixes #8898. Fixes #61542. Change-Id: Ieb303b6de1fb3527d3256135151a9e983f3c27e6 Reviewed-on: https://go-review.googlesource.com/c/go/+/512355 Reviewed-by: Austin Clements <austin@google.com> LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com> Auto-Submit: Russ Cox <rsc@golang.org>
2024-02-14 18:36:47 -07:00
Go 1.23 changed the mode bits reported by [`os.Lstat`](/pkg/os#Lstat) and [`os.Stat`](/pkg/os#Stat)
for reparse points, which can be controlled with the `winsymlink` setting.
As of Go 1.23 (`winsymlink=1`), mount points no longer have [`os.ModeSymlink`](/pkg/os#ModeSymlink)
set, and reparse points that are not symlinks, Unix sockets, or dedup files now
always have [`os.ModeIrregular`](/pkg/os#ModeIrregular) set. As a result of these changes,
[`filepath.EvalSymlinks`](/pkg/path/filepath#EvalSymlinks) no longer evaluates
mount points, which was a source of many inconsistencies and bugs.
At previous versions (`winsymlink=0`), mount points are treated as symlinks,
and other reparse points with non-default [`os.ModeType`](/pkg/os#ModeType) bits
(such as [`os.ModeDir`](/pkg/os#ModeDir)) do not have the `ModeIrregular` bit set.
Go 1.23 changed [`os.Readlink`](/pkg/os#Readlink) and [`filepath.EvalSymlinks`](/pkg/path/filepath#EvalSymlinks)
to avoid trying to normalize volumes to drive letters, which was not always even possible.
This behavior is controlled by the `winreadlinkvolume` setting.
For Go 1.23, it defaults to `winreadlinkvolume=1`.
Previous versions default to `winreadlinkvolume=0`.
Go 1.23 enabled the experimental post-quantum key exchange mechanism
X25519Kyber768Draft00 by default. The default can be reverted using the
[`tlskyber` setting](/pkg/crypto/tls/#Config.CurvePreferences).
Go 1.23 changed the behavior of
[crypto/x509.ParseCertificate](/pkg/crypto/x509/#ParseCertificate) to reject
serial numbers that are negative. This change can be reverted with
the [`x509negativeserial` setting](/pkg/crypto/x509/#ParseCertificate).
Go 1.23 changed the behavior of
[crypto/x509.ParseCertificate](/pkg/crypto/x509/#ParseCertificate) to reject
serial numbers that are longer than 20 octets. This change can be reverted with
the [`x509seriallength` setting](/pkg/crypto/x509/#ParseCertificate).
Go 1.23 re-enabled support in html/template for ECMAScript 6 template literals by default.
The [`jstmpllitinterp` setting](/pkg/html/template#hdr-Security_Model) no longer has
any effect.
Go 1.23 changed the default TLS cipher suites used by clients and servers when
not explicitly configured, removing 3DES cipher suites. The default can be reverted
using the [`tls3des` setting](/pkg/crypto/tls/#Config.CipherSuites).
Go 1.23 changed the behavior of [`tls.X509KeyPair`](/pkg/crypto/tls#X509KeyPair)
and [`tls.LoadX509KeyPair`](/pkg/crypto/tls#LoadX509KeyPair) to populate the
Leaf field of the returned [`tls.Certificate`](/pkg/crypto/tls#Certificate).
This behavior is controlled by the `x509keypairleaf` setting. For Go 1.23, it
defaults to `x509keypairleaf=1`. Previous versions default to
`x509keypairleaf=0`.
### Go 1.22
Go 1.22 adds a configurable limit to control the maximum acceptable RSA key size
that can be used in TLS handshakes, controlled by the [`tlsmaxrsasize` setting](/pkg/crypto/tls#Conn.Handshake).
The default is tlsmaxrsasize=8192, limiting RSA to 8192-bit keys. To avoid
denial of service attacks, this setting and default was backported to Go
1.19.13, Go 1.20.8, and Go 1.21.1.
Go 1.22 made it an error for a request or response read by a net/http
client or server to have an empty Content-Length header.
This behavior is controlled by the `httplaxcontentlength` setting.
Go 1.22 changed the behavior of ServeMux to accept extended
patterns and unescape both patterns and request paths by segment.
This behavior can be controlled by the
[`httpmuxgo121` setting](/pkg/net/http/#ServeMux).
Go 1.22 added the [Alias type](/pkg/go/types#Alias) to [go/types](/pkg/go/types)
for the explicit representation of [type aliases](/ref/spec#Type_declarations).
Whether the type checker produces `Alias` types or not is controlled by the
[`gotypesalias` setting](/pkg/go/types#Alias).
For Go 1.22 it defaults to `gotypesalias=0`.
For Go 1.23, `gotypesalias=1` will become the default.
This setting will be removed in a future release, Go 1.27 at the earliest.
Go 1.22 changed the default minimum TLS version supported by both servers
and clients to TLS 1.2. The default can be reverted to TLS 1.0 using the
[`tls10server` setting](/pkg/crypto/tls/#Config).
Go 1.22 changed the default TLS cipher suites used by clients and servers when
not explicitly configured, removing the cipher suites which used RSA based key
exchange. The default can be reverted using the [`tlsrsakex` setting](/pkg/crypto/tls/#Config).
Go 1.22 disabled
[`ConnectionState.ExportKeyingMaterial`](/pkg/crypto/tls/#ConnectionState.ExportKeyingMaterial)
when the connection supports neither TLS 1.3 nor Extended Master Secret
(implemented in Go 1.21). It can be reenabled with the [`tlsunsafeekm`
setting](/pkg/crypto/tls/#ConnectionState.ExportKeyingMaterial).
Go 1.22 changed how the runtime interacts with transparent huge pages on Linux.
In particular, a common default Linux kernel configuration can result in
significant memory overheads, and Go 1.22 no longer works around this default.
To work around this issue without adjusting kernel settings, transparent huge
pages can be disabled for Go memory with the
[`disablethp` setting](/pkg/runtime#hdr-Environment_Variable).
This behavior was backported to Go 1.21.1, but the setting is only available
starting with Go 1.21.6.
This setting may be removed in a future release, and users impacted by this issue
should adjust their Linux configuration according to the recommendations in the
[GC guide](/doc/gc-guide#Linux_transparent_huge_pages), or switch to a Linux
distribution that disables transparent huge pages altogether.
Go 1.22 added contention on runtime-internal locks to the [`mutex`
profile](/pkg/runtime/pprof#Profile). Contention on these locks is always
reported at `runtime._LostContendedRuntimeLock`. Complete stack traces of
runtime locks can be enabled with the [`runtimecontentionstacks`
setting](/pkg/runtime#hdr-Environment_Variable). These stack traces have
non-standard semantics, see setting documentation for details.
Go 1.22 added a new [`crypto/x509.Certificate`](/pkg/crypto/x509/#Certificate)
field, [`Policies`](/pkg/crypto/x509/#Certificate.Policies), which supports
certificate policy OIDs with components larger than 31 bits. By default this
field is only used during parsing, when it is populated with policy OIDs, but
not used during marshaling. It can be used to marshal these larger OIDs, instead
of the existing PolicyIdentifiers field, by using the
[`x509usepolicies` setting.](/pkg/crypto/x509/#CreateCertificate).
### Go 1.21
Go 1.21 made it a run-time error to call `panic` with a nil interface value,
controlled by the [`panicnil` setting](/pkg/builtin/#panic).
Go 1.21 made it an error for html/template actions to appear inside of an ECMAScript 6
template literal, controlled by the
[`jstmpllitinterp` setting](/pkg/html/template#hdr-Security_Model).
This behavior was backported to Go 1.19.8+ and Go 1.20.3+.
Go 1.21 introduced a limit on the maximum number of MIME headers and multipart
forms, controlled by the
[`multipartmaxheaders` and `multipartmaxparts` settings](/pkg/mime/multipart#hdr-Limits)
respectively.
This behavior was backported to Go 1.19.8+ and Go 1.20.3+.
net: mptcp: force using MPTCP with GODEBUG When adding MPTCP support to address the proposal #56539, I missed the GODEBUG setting from Russ Cox's plan: I am inclined to say that we add MPTCP as an opt-in for a release or two, and then make it opt-out. There should be a GODEBUG setting (...) See: https://github.com/golang/go/issues/56539#issuecomment-1309294637 Thanks to andrius4669 for having reported this issue to me. It makes sense to have this GODEBUG setting not to have to modify applications to use MPTCP (if available). It can then be useful to estimate the impact in case we want to switch from opt-in to opt-out later. The MPTCP E2E test has been modified to make sure we can enable MPTCP either via the source code like it was already the case before or with this environment variable: GODEBUG=multipathtcp=1 The documentation has been adapted accordingly. I don't know if it is too late for Go 1.21 but I had to put a version in the documentation. The modification is small, the risk seems low and this was supposed to be there from the beginning according to Russ Cox's specifications. It can also be backported or only be present in the future v1.22 if it is easier. Note: I didn't re-open #56539 or open a new one. It is not clear to me what I should do in this case. Fixes #56539 Change-Id: I9201f4dc0b99e3643075a34c7032a95528c48fa0 Reviewed-on: https://go-review.googlesource.com/c/go/+/507375 Reviewed-by: Cherry Mui <cherryyz@google.com> Auto-Submit: Ian Lance Taylor <iant@golang.org> Run-TryBot: Ian Lance Taylor <iant@golang.org> TryBot-Result: Gopher Robot <gobot@golang.org> Reviewed-by: Damien Neil <dneil@google.com>
2023-06-30 09:24:57 -06:00
Go 1.21 adds the support of Multipath TCP but it is only used if the application
explicitly asked for it. This behavior can be controlled by the
[`multipathtcp` setting](/pkg/net#Dialer.SetMultipathTCP).
There is no plan to remove any of these settings.
### Go 1.20
Go 1.20 introduced support for rejecting insecure paths in tar and zip archives,
controlled by the [`tarinsecurepath` setting](/pkg/archive/tar/#Reader.Next)
and the [`zipinsecurepath` setting](/pkg/archive/zip/#NewReader).
These default to `tarinsecurepath=1` and `zipinsecurepath=1`,
preserving the behavior of earlier versions of Go.
A future version of Go may change the defaults to
`tarinsecurepath=0` and `zipinsecurepath=0`.
Go 1.20 introduced automatic seeding of the
[`math/rand`](/pkg/math/rand) global random number generator,
controlled by the [`randautoseed` setting](/pkg/math/rand/#Seed).
Go 1.20 introduced the concept of fallback roots for use during certificate verification,
controlled by the [`x509usefallbackroots` setting](/pkg/crypto/x509/#SetFallbackRoots).
Go 1.20 removed the preinstalled `.a` files for the standard library
from the Go distribution.
Installations now build and cache the standard library like
packages in other modules.
The [`installgoroot` setting](/cmd/go#hdr-Compile_and_install_packages_and_dependencies)
restores the installation and use of preinstalled `.a` files.
There is no plan to remove any of these settings.
### Go 1.19
Go 1.19 made it an error for path lookups to resolve to binaries in the current directory,
controlled by the [`execerrdot` setting](/pkg/os/exec#hdr-Executables_in_the_current_directory).
There is no plan to remove this setting.
### Go 1.18
Go 1.18 removed support for SHA1 in most X.509 certificates,
controlled by the [`x509sha1` setting](/pkg/crypto/x509#InsecureAlgorithmError).
This setting will be removed in a future release, Go 1.22 at the earliest.
### Go 1.10
Go 1.10 changed how build caching worked and added test caching, along
with the [`gocacheverify`, `gocachehash`, and `gocachetest` settings](/cmd/go/#hdr-Build_and_test_caching).
There is no plan to remove these settings.
### Go 1.6
Go 1.6 introduced transparent support for HTTP/2,
controlled by the [`http2client`, `http2server`, and `http2debug` settings](/pkg/net/http/#hdr-HTTP_2).
There is no plan to remove these settings.
### Go 1.5
Go 1.5 introduced a pure Go DNS resolver,
controlled by the [`netdns` setting](/pkg/net/#hdr-Name_Resolution).
There is no plan to remove this setting.