Update pre-null-safety spec with new null safety sections #2023
Conversation
|
An update fixing a few issues with interface types is on the way, will upload in half an hour, approx. ;-) |
|
Done. |
eernstg
force-pushed
the
specify_null_safety_new_sections_dec21
branch
from
3829332
to
f5cb6da
Compare
|
Visit the preview URL for this PR (updated for commit 6cd9296): https://dart-specification--pr2023-specify-null-safety-56htdccy.web.app (expires Mon, 11 Jul 2022 14:42:40 GMT) 🔥 via Firebase Hosting GitHub Action 🌎 Sign: 6941ecd630c4f067ff3d02708a45ae0f0a42b88a |
eernstg
force-pushed
the
specify_null_safety_new_sections_dec21
branch
from
f5cb6da
to
24422fc
Compare
eernstg
force-pushed
the
specify_null_safety_new_sections_dec21
branch
from
24422fc
to
0fa020d
Compare
|
I skimmed over the changes and created the following checklist. I plan to review the text in parts and cross them out here when I'm done. The checklist itself may change in the future in case I missed anything.
|
|
I checked the part "Rules of Inference", in the green "after" column the lines range from 20728 to 20889. It looks good to me. EDIT: In my review I assumed that the requirement for the bounds of two function types is to be in the mutual-subtypes relationship. After a conversation with Erik I realized that the spec needs to be updated to mean that. |
|
I reviewed the part "Subsequent commentary" to the subtype algorithm, in the green "after" column the lines range from 20889 to 21291. I have two drive-by questions to the surrounding text. Otherwise, the part LGTM. |
|
I checked the part "Type Nullability", in the green "after" column the lines range from 21291 to 21414. I have left some comments, but otherwise LGTM. |
|
I checked the part "Top merge", in the green "after" column the lines range from 21414 to 21494. LGTM with one comment. |
|
I checked the part "IsTopType", in the green "after" column the lines range from 21494 to 21539. It LGTM. |
|
I checked the part "Moretop, morebottom", in the green "after" column the lines range from 21539 to 21582. LGTM. |
|
I checked the part "Type normalization", in the green "after" column the lines range from 21582 to 21737. LGTM with one question. |
|
I checked the part "Canonical syntax of types", in the green "after" column the lines range from 21737 to 21931. LGTM. |
| \end{dartCode} | ||
|
|
||
| \commentary{% | ||
| No \emph{least} upper bound exists for \code{A} and \code{B}, |
There was a problem hiding this comment.
As I understand it, with the exception of a top type?
There was a problem hiding this comment.
This particular sentence refers to the real, mathematical least upper bound in a set (Dart types) with an ordering relation (subtyping). The usage of 'least upper bound' in older versions of the spec is misleading, because it was never a least upper bound.
So we're looking for a type which is a supertype of both A and B, and it must be the least element among all the types which are supertypes of both A and B. It can't be I because we don't have I <: J; similarly it can't be J. But it also can't be a top type, because we don't have Top <: I (or J, or a bunch of other types).
So I think the claim is correct.
There was a problem hiding this comment.
But it also can't be a top type, because we don't have
Top <: I(orJ, or a bunch of other types).
I don't think I understand this argument. Shouldn't we have I <: Top and J <: Top (in the reverse order) in order to claim Top as the upper bound for both I and J?
Additionally, I think declarations class I {} and class J {} imply class I extends Object {} respectively class J extends Object {}, which makes Object an upper bound for A and B.
There was a problem hiding this comment.
True, Object is an upper bound for each of those classes, but that makes no difference: The least upper bound (mathematically speaking) of a set S of elements in an ordered set {v1, .. vk} is an element v from S that satisfies (1) vj <: v, for all j, and (v1 <: other && .. && vk <: other) => v <: other. It's the latter part that fails, so Top and Object aren't 'least', they are just upper bounds.
|
I checked the part "Types bounded by types", in the green "after" column the lines range from 22983 to 23015. LGTM with one small question. |
|
I checked the parts "Class building types", "Interface types", "Intersection types", "Never", "Null", "Type", "Function", "dynamic", in the green "after" column the lines range from 23015 to 24140. LGTM with some comments. |
specification/dartLangSpec.tex
Outdated
| \LMHash{}% | ||
| A check of the form \code{$v$\,\,!=\,\,\NULL}, | ||
| \code{\NULL\,\,!=\,\,$v$}, | ||
| or \code{$v$\,\,\IS\,\,$T$} |
There was a problem hiding this comment.
Is it true for every type T? What about the case of T = S??
There was a problem hiding this comment.
I changed the text to follow the null safety spec precisely about the rule such that we don't need to worry about unintended novelties.
However, v is T where the static type of v is T? is oddly specific, and it certainly doesn't cover the precise behavior of the CFE nor the analyzer.
So we would presumably need to find a way to describe the current behavior of the tools in more detail, such that they can continue to handle cases like X extends int? and still be in sync with the spec. I'll create an issue for that.
There was a problem hiding this comment.
I made some changes to the text according to the response in #2258.
|
I checked the part "Type inference", in the green "after" column the lines range from 24140 to 24613. LGTM with some minor questions. |
specification/dartLangSpec.tex
Outdated
| \item $B_{0i}[Z_0/X_0, \ldots, Z_k/X_k]$ and $B_{1i}[Z_0/Y_0, \ldots, Z_k/Y_k]$ | ||
| have the same canonical syntax, for $i \in 0 .. k$. |
There was a problem hiding this comment.
Should we ask for a weaker requirement: the bounds should be mutual subtypes or should be equal if all top types treated as being equal to each other and the bottom types treated as being equal to each other?
There was a problem hiding this comment.
Yes! The requirement should be 'mutual subtypes', and it's been updated now. I overlooked the change were === in the null safety feature spec was changed to mean 'mutual subtypes', and that's the reason why I kept the old text because === was still there. That shows the power of change propagation! ;-)
|
I checked the part "Appendix. Algorithmic subtyping", in the green "after" column the lines range fro 24613 to 24847. LGTM with some question. That concludes my initial pass over the PR. |
specification/dartLangSpec.tex
Outdated
| \LMHash{}% | ||
| A check of the form \code{$v$\,\,!=\,\,\NULL}, | ||
| \code{\NULL\,\,!=\,\,$v$}, | ||
| or \code{$v$\,\,\IS\,\,$T$} |
There was a problem hiding this comment.
I changed the text to follow the null safety spec precisely about the rule such that we don't need to worry about unintended novelties.
However, v is T where the static type of v is T? is oddly specific, and it certainly doesn't cover the precise behavior of the CFE nor the analyzer.
So we would presumably need to find a way to describe the current behavior of the tools in more detail, such that they can continue to handle cases like X extends int? and still be in sync with the spec. I'll create an issue for that.
specification/dartLangSpec.tex
Outdated
| \LMHash{}% | ||
| A check of the form \code{$v$\,\,!=\,\,\NULL}, | ||
| \code{\NULL\,\,!=\,\,$v$}, | ||
| or \code{$v$\,\,\IS\,\,$T$} |
specification/dartLangSpec.tex
Outdated
| \LMHash{}% | ||
| A check of the form \code{$v$\,\,!=\,\,\NULL}, | ||
| \code{\NULL\,\,!=\,\,$v$}, | ||
| or \code{$v$\,\,\IS\,\,$T$} |
There was a problem hiding this comment.
I made some changes to the text according to the response in #2258.
specification/dartLangSpec.tex
Outdated
| \item $B_{0i}[Z_0/X_0, \ldots, Z_k/X_k]$ and $B_{1i}[Z_0/Y_0, \ldots, Z_k/Y_k]$ | ||
| have the same canonical syntax, for $i \in 0 .. k$. |
There was a problem hiding this comment.
Yes! The requirement should be 'mutual subtypes', and it's been updated now. I overlooked the change were === in the null safety feature spec was changed to mean 'mutual subtypes', and that's the reason why I kept the old text because === was still there. That shows the power of change propagation! ;-)
eernstg
force-pushed
the
specify_null_safety_new_sections_dec21
branch
from
b9279cb
to
6cd9296
Compare
|
The PR was rebased on the current |
eernstg
force-pushed
the
specify_null_safety_new_sections_dec21
branch
from
6cd9296
to
66420dd
Compare
eernstg
force-pushed
the
specify_null_safety_new_sections_dec21
branch
from
185037e
to
ddd09ec
Compare
eernstg
force-pushed
the
specify_null_safety_new_sections_dec21
branch
from
ddd09ec
to
6c9de7a
Compare
eernstg
force-pushed
the
specify_null_safety_new_sections_dec21
branch
from
ea307ec
to
430eaf5
Compare
eernstg
force-pushed
the
specify_null_safety_new_sections_dec21
branch
from
430eaf5
to
fb18232
Compare
This PR was created by deleting the sections from
Subtypesto (but not including)Referenceplus the appendixAlgorithmic Subtyping, and inserting the sectionSubtypesfrom the null safety update branch plus all the following sections until (but not including)Reference, plus the new version of the appendixAlgorithmic Subtyping.The point is that this PR introduces a large chunk of the new material related to null safety which is introduced into the language specification, and it does not include all the small changes to sections that are mostly the same as before (those changes will be submitted as other PRs later on). In other words, the text before
Subtypingis unchanged (apart from the fixes mentioned below), and references from the new material to any location beforeSubtypingmay thus very well refer to obsolete language.A couple of other updates were necessary in order to make the result LaTeX'able:
dart.stywas updated to be the version used with the null safety PR (the new version includes a number of declarations that are used by the material about null safety).dartLangSpec.texwas updated to use\Deltarather than\Gammato denote a type variable environment (that's the common notation; we previously used\Gamma, but that's typically a "term" variable environment, so\Deltawould be less confusing for readers who also read research papers involving type systems).\RawFunctionTypeNamedwas updated to allow for therequiredmodifier; a couple of changes were necessary in sections beforeSubtypingin order to make dartLangSpec.tex LaTeX'able again, but this only involved adding the new LaTeX macro argument{r}, further updates will take place later on (this update just updates stuff fromSubtypingand out).\diamondto\triangle, because\triangleis now also used over=in order to indicate that a certain term of the formA = BdefinesA.\ominussymbol, similar to the symbolic traffic sign meaning "No entrance"), this symbol is used in the new material, that is, nowhere before sectionSubtyping; for instance, it occurs in sectionType Null.Unfortunately, the diff algorithm in git (and, probably, almost any diff algorithm) does not quite understand that the operation performed here is "delete two big chunks of text and insert replacements that are mostly new material", but it does seem to get on track after a while.