-
Notifications
You must be signed in to change notification settings - Fork 517
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Feedback on swapping out the chrono::Duration type #1268
Comments
Going to close this. Seems like we're good. Feedback is still welcome! |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Please leave your feedback here. See full context below for reference.
In the 0.4.30 release, we have decided to swap out the
chrono::Duration
type (which has been a re-export of time 0.1Duration
type) with our own defition, which exposes a strict superset of thetime::Duration
API. This helps avoid warnings about the CVE-2020-26235 and RUSTSEC-2020-0071 advisories for downstream users and allows us to improve theDuration
API going forward.While this is technically a SemVer-breaking change, we expect the risk of downstream users experiencing actual incompatibility to be exceedingly limited (see our analysis of public code using a crater-like experiment), and not enough justification for the large ecosystem churn of a 0.5 release.
Relation between chrono and time 0.1
Rust first had a
time
module added tostd
in its 0.7 release. It later moved tolibextra
, and then to alibtime
library shipped alongside the standard library. In 2014 work on chrono started in order to provide a full-featured date and time library in Rust. Some improvements from chrono made it into the standard library; notably,chrono::Duration
was included asstd::time::Duration
(rust#15934) in 2014.In preparation of Rust 1.0 at the end of 2014
libtime
was moved out of the Rust distro and into thetime
crate to eventually be redesigned (rust#18832, rust#18858), like thenum
andrand
crates. Of course chrono kept its dependency on thistime
crate.time
started re-exportingstd::time::Duration
during this period. Later, the standard library was changed to have a more limited unsignedDuration
type (rust#24920, RFC 1040), while thetime
crate kept the full functionality withtime::Duration
.time::Duration
had been a part of chrono's public API.By 2016
time
0.1 lived under therust-lang-deprecated
organisation and was not actively maintained (time#136). chrono absorbed the platform functionality andDuration
type of thetime
crate in chrono#478 (the work started in chrono#286). In order to preserve compatibility with downstream crates depending ontime
andchrono
sharing aDuration
type, chrono kept depending on time 0.1. chrono offered the option to opt out of thetime
dependency by disabling theoldtime
feature (swapping it out for an effectively similar chrono type). In 2019, @jhpratt took over maintenance on thetime
crate and released what amounts to a new crate astime
0.2.Security advisories
In November of 2020 CVE-2020-26235 and RUSTSEC-2020-0071 were opened against the
time
crate. @quininer had found that calls tolocaltime_r
may be unsound (chrono#499). Eventually, almost a year later, this was also made into a security advisory against chrono as RUSTSEC-2020-0159, which had platform code similar totime
.On Unix-like systems a process is given a timezone id or description via the
TZ
environment variable. We need this timezone data to calculate the current local time from a value that is in UTC, such as the time from the system clock.time
0.1 and chrono used the POSIX functionlocaltime_r
to do the conversion to local time, which reads theTZ
variable.Rust assumes the environment to be writable and uses locks to access it from multiple threads. Some other programming languages and libraries use similar locking strategies, but these are typically not shared across languages. More importantly, POSIX declares modifying the environment in a multi-threaded process as unsafe, and
getenv
in libc can't be changed to take a lock because it returns a pointer to the data (see rust#27970 for more discussion).Since version 4.20 chrono no longer uses
localtime_r
, instead using Rust code to query the timezone (from theTZ
variable or viaiana-time-zone
as a fallback) and work with data from the system timezone database directly. The code for this was forked from the tz-rs crate by @x-hgg-x. As such, chrono now respects the Rust lock when reading theTZ
environment variable. In general, code should avoid modifying the environment.Removing time 0.1
Because time 0.1 has been unmaintained for years, however, the security advisory mentioned above has not been addressed. While chrono maintainers were careful not to break backwards compatibility with the
time::Duration
type, there has been a long stream of issues from users inquiring about the time 0.1 dependency with the vulnerability. We investigated the potential breakage of removing the time 0.1 dependency in [chrono#1095] using a crater-like experiment and determined that the potential for breaking (public) dependencies is very low. We reached out to those few crates that did still depend on compatibility with time 0.1.As such, for chrono 0.4.30 we have decided to swap out the time 0.1
Duration
implementation for a local one that will offer a strict superset of the existing API going forward. This will prevent most downstream users from being affected by the security vulnerability in time 0.1 while minimizing the ecosystem impact of semver-incompatible version churn.The text was updated successfully, but these errors were encountered: