rfc(374): The jsii compiler to follow TypeScript versioning #373
Conversation
This PR proposes a new way to model the `aws/jsii` project in order to alleviate friction in making significant changes to the platform, without compromising on key elements of the developer experience, or risking to unnecessarily fragment the ecosystem.
RomainMuller
force-pushed
the
rmuller/jsii-version-unlock
branch
from
f15e172 to
d9924ca
Compare
RomainMuller
changed the title
text/0374-jsii-modular.md
Outdated
| maintain backwards compatibility between individual packages, and reduce the | ||
| risk of creating codependent packages (different repositories need to be | ||
| released independenlty, so dependencies can only flow in an asyclic manner). |
There was a problem hiding this comment.
Don't monorepos provide the same or some of these assurances? For eg, I believe you can't have cyclic dependencies within packages of a monorepo.
It's not clear to me (from the rfc) why coupling jsii versions and breaking off monorepo are related.
By not breaking off the monorepo, some of the concerns you've listed in the next question can be avoided while still decoupling jsii release versions.
There was a problem hiding this comment.
I am missing a deeper discussion on compatibility boundaries between the tools.
- The .jsii manifest is a boundary between the compiler and pacmak (and jsii-relect) (do we really need those two to be separate tools? Maybe we can merge them into a single tool).
- There is also a compatibility boundary between the runtime and the generated code. If two modules use different versions of pacmak, do they need two runtimes?
In my mind having a formal compatibility model is the first thing we must finalize before we are able to clearly break this coupling.
On the other hand, there are some low-hanging fruit that we can easily extract from the main repo (e.g. superchain).
RomainMuller
force-pushed
the
rmuller/jsii-version-unlock
branch
from
e255213 to
5d1e3d1
Compare
RomainMuller
changed the title
| * **Optional:** Separate the `jsii` compiler from the rest of packages in the | ||
| jsii toolchain into separate repositories. This would make it easier to | ||
| release the `jsii` compiler separately from other parts of the toolchain that | ||
| follow a different versioning scheme. |
There was a problem hiding this comment.
Possible projen use case?
| This is not a breaking change in the traditional sense, however this is a | ||
| significant change in versioning schemes used for the `jsii` compiler, and a | ||
| departure from the [AWS SDKs and Tools maintenance policy][aws-policy]. | ||
|
|
||
| As such, this warrants ample communication with our customer base to avoid they | ||
| are caught by surprise, or have broken expectations. |
There was a problem hiding this comment.
We should be clear: it either is a breaking change or it isn't. If customers have to change their code (I think they might have to since newer TypeScript versions introduce breaking changes) then it's a breaking change, I think. If any migration does not require customers to change any of their code, then it's not a breaking change. WDYT?
There was a problem hiding this comment.
What I want to say here is that we are making a breaking change... to our versioning scheme (SemVer -> TypeScript-Ver).
Customers will have to update their code when they move from jsii@1.x to jsii@4.7.x due to language specification changes, but the real think they need to be very carful about is to update from a caret dependency jsii@^1.0.0 to a tilde dependency jsii@~4.7.0.
The level/duration of maintenance for future versions will also be different, and we should definitely encourage users to closely track latest as much as possible.
Co-authored-by: Momo Kornher <mail@moritzkornher.de>
Co-authored-by: Momo Kornher <mail@moritzkornher.de>
text/0374-jsii-ts-version.md
Outdated
| others) are not affected by this change and will continue to follow [semantic | ||
| versioning][semver] as they currently do. The bindings generated by |
There was a problem hiding this comment.
Does this document say anything about the backwards compatibility guarantees of these tools?
For example, we could be MV-bumping on every release, and proudly claim "we are following semver", yet our customers would still hate us.
There was a problem hiding this comment.
It does in the sense that it has provisions around how you shouldn't be forced to upgrade your dependency on a jsii tool because one of your dependencies or dependents has upgraded already (so every new line must be able to operate on output of previous AND future versions).
| currently active [TypeScript] release line. | ||
|
|
||
| Critical bug fixes and security patches will be provided for previous release | ||
| lines for 12 months, which covers 3 or 4 previous lines. Features will not be |
There was a problem hiding this comment.
In the current state of things, bug-and-security maintenance for a compiler line does not require a ton of work (few bug reports against the compiler), and so I am assuming 1 year of maintenance is not an issue.
That said, this is a line in the sand and we can absolutely discuss where it should be put.
| versioning][semver], this implies `jsii` and `jsii-rosetta` also need to stop | ||
| conforming to it. | ||
|
|
||
| ### Why should we _not_ do this? |
There was a problem hiding this comment.
Another "why should we not do this": customers will need to potentially change their code to get access to bug fixes and features in new jsii releases.
There was a problem hiding this comment.
to get access to bug fixes
Only if they are currently on a compiler release that is end-of-life, which... I guess is the same with all end-of-life software.
and features
This is by design!
There was a problem hiding this comment.
Only if they are currently on a compiler release that is end-of-life, which... I guess is the same with all end-of-life software.
I might have missed this, but are you backporting all bugfixes to all versions?
and features
This is by design!
Well. Yes for TSC features, but also for jsii-related features? Let's say we add a "named type unions" feature to jsii... would that feature also be backported to all 4 current release lines? Or just be added to the latest?
That's quite a heavy maintenance burden you would be taking on if yes...
There was a problem hiding this comment.
I might have missed this, but are you backporting all bugfixes to all versions?
There is a whole section on the maintenance policy that states we will fix bugs in releases for 1 year after they cease to be "current" (I have added alternate scenarios there -- one for 6 months instead of 1 year, and one for "only current + previous" which is what Go does). Sooo technically not "all versions", but "some" for sure.
Let's say we add a "named type unions" feature to jsii...
That is certainly an interesting case, because such a thing does not exist in the current type model... We'd either need to design a strategy that allows us to add that feature in a strictly backwards compatible manner (sounds hard if even possible w/o voiding the feature entirely), or we might need to back-port that feature on non-EOL compilers (and yes - in this case that sounds like more work than I'd like to sign up for, if we uphold the 12 months policy).
There was a problem hiding this comment.
I added a blurb in "why not" about this as you suggested. Not sure this is a deal breaker here, but certainly should inform our final decision on the maintenance/support policy.
There was a problem hiding this comment.
I'd prefer we didn't backport changes like that for a whole year. It was a ton of maintenance work for the oncall when we did it for CDK...it made the oncall's life much easier when CDK v1 finally hit EOL.
There was a problem hiding this comment.
Found some tiny typos while reading through your RFC, @RomainMuller 🔍😄
I really like this proposal!
It was interesting to read as matching a TypeScript version with JSIIs major.minor sounds like a solution to a quite similar problem we're facing with pre-built provider bindings in CDKTF which depend on the CDKTF version and the underlying Terraform Provider version (also TIL about cdk8s approach here).
Co-authored-by: Ansgar Mertens <mertens.ansgar@gmail.com>
Co-authored-by: Ansgar Mertens <mertens.ansgar@gmail.com>
Co-authored-by: Ansgar Mertens <mertens.ansgar@gmail.com>
| versioning][semver], this implies `jsii` and `jsii-rosetta` also need to stop | ||
| conforming to it. | ||
|
|
||
| ### Why should we _not_ do this? |
There was a problem hiding this comment.
I'd prefer we didn't backport changes like that for a whole year. It was a ton of maintenance work for the oncall when we did it for CDK...it made the oncall's life much easier when CDK v1 finally hit EOL.
| line, including a planned timeline for the initial release of the `v4.9` line, | ||
| and language about the departure from [SemVer]. | ||
|
|
||
| * **Optional:** Separate the `jsii` compiler from the rest of packages in the |
There was a problem hiding this comment.
This is worthwhile. I think it's confusing if jsii follows different versioning than jsii-pacmak when they live in the same repo.
|
|
||
| - Make the `github.com/aws/jsii` repository by a single-package repository. | ||
|
|
||
| * Update the *release* automation to use the [TypeScript] compiler version's |
There was a problem hiding this comment.
does this mean that new jsii versions can only be made when there are new TSC releases? Should we add a new dot to the release for any critical regression fixes, so we have a
major.minor.patch.jsiiPatch scheme?
There was a problem hiding this comment.
I'm pretty sure npm will not accept that version scheme.
Concretely, we are free to make new versions whenever we want under the new scheme, but they'll all be "patch" versions (introduction of new features, bug fixes, etc... all those as patch versions).
We can however only make BREAKING changes in a version when a new TypeScript compiler version is released (this happens once per quarter, which I believe is frequent enough).
Co-authored-by: Calvin Combs <66279577+comcalvi@users.noreply.github.com>
This RFC #374 proposes to change the versioning scheme of the
jsiicompiler to stopconforming to semantic versioning, and instead use the
major.minorversion of the TypeScript compiler it is built on.
By submitting this pull request, I confirm that my contribution is made under
the terms of the Apache-2.0 license