Skip to content

Latest commit

 

History

History
638 lines (509 loc) · 34.2 KB

RELEASES.md

File metadata and controls

638 lines (509 loc) · 34.2 KB
Raw

Version 1.6 (2024-04-25)

Nickel 1.6 is a maintenance release including several bug fixes and improvements, in particular around the features introduced in Nickel 1.5 (enum variants and background evaluation in the LSP).

Core language

  • Extend merge to enum variants by @yannham in #1862
  • [Fix] Allow multiple underscore to start identifiers by @yannham in #1884

Stdlib

  • Add std.string.find_all by @fuzzypixelz in #1870
  • Add empty optional fields-aware record operator variants by @yannham in #1876

Documentation

  • modular-configurations.md: to_lower -> lowercase by @Jasha10 in #1857
  • manual/modular-configurations.md: add argument to std.string.join by @Jasha10 in #1859
  • manual/syntax.md: minor typo by @Jasha10 in #1860

LSP

  • Improve diagnostic location in nls by @jneem in #1856
  • Propagate pending array contracts in permissive_eval by @jneem in #1854
  • Don't leak memory in background eval by @jneem in #1869
  • Add a recursion limit to background evaluation by @jneem in #1878
  • Dedup diagnostics by @jneem in #1883
  • Extend the symbol range to include the rhs by @jneem in #1887
  • Leverage function contract information by @yannham in #1888
  • Fix LSP not showing type signature in untyped code by @yannham in #1889

Tooling

  • Markdown documentation generation: do not insert line breaks by @yannham in #1879
  • [Fix] Nickel doc: fix missing newline in markdown output by @yannham in #1880
  • Fix infinite recursion in doc symbols. by @jneem in #1881

Fixes

  • [Fix] Polymorphic field typechecking by @yannham in #1872
  • Force enum payloads by @jneem in #1890

Version 1.5 (2024-03-12)

Nickel 1.5 is a major release (albeit not literally), with new core language features and improved LSP. In particular, Nickel 1.5 introduces:

  • Full blown pattern matching. Patterns were previously restricted to destructuring let-bindings. They can now be used within match expressions as well (which only supported simple enum tags before).
  • Enum variants. Enum variants are a new language constructs which are enum tags with associated data (they are applied to an argument). They can be seen as form of algebraic data types (ADT). Patterns, enum types, typechecking and other parts of the language are extended accordingly to support them.
  • Background evaluation in the LSP. The LSP now performs evaluation of the current document in the background to report evaluation errors directly in your editor, and in particular contract errors.
  • A new section of the manual on writing modular configurations.

Core language

  • Allow metadata keywords in field position by @yannham in #1768

  • Support other bases than decimal for num literals by @yannham in #1798

  • Enum variants:

    • Structural ADTs: first step by @yannham in #1770
    • Implement equality on ADTs by @yannham in #1787
    • Fix laziness closurization bug, add support for ADTs by @yannham in #1789
    • ADT destructuring by @yannham in #1812
    • Enum tag destructuring by @yannham in #1813
    • Implement ADT contracts by @yannham in #1821
    • Introduce application syntax for ADTs by @yannham in #1825
    • Relax enum row conflicts by @yannham in #1831
    • Properly force enum variants by @yannham in #1835
    • Fix enum contract stripping unwrapping variant argument by @yannham in #1833

  • Pattern matching:

    • [Refactor] Pattern matching by @yannham in #1799
    • Record pattern compilation by @yannham in #1816
    • Enum pattern compilation by @yannham in #1817
    • [Refactor] Pattern positions by @yannham in #1819
    • Full pattern matching by @yannham in #1820
    • Support pattern contracts in match statement by @yannham in #1823
    • Fix the semantics of default values in patterns by @yannham in #1826
    • Specialized pattern compilation for enum tags by @yannham in #1846

Stdlib

  • Add array.zip_with and array.map_with_index to the standard library by @vkleen in #1797
  • fixed std.array.split_at behavior at right boundary. by @suimong in #1803
  • Update stdlib for ADTs by @yannham in #1822

Tooling

  • LSP: Add "goto definition" support for the import term by @jneem in #1756
  • LSP: Add support for NICKEL_IMPORT_PATH environment variable @jneem in #1795
  • LSP: Improved reference-finding by @jneem in #1800
  • LSP: rename action by @jneem in #1811
  • LSP: evaluation in the background by @jneem in #1814
  • LSP: Improve document symbols by @jneem in #1848
  • Add more spacing to contract error messages by @yannham in #1767
  • Proper error message for non exhaustive match by @yannham in #1772
  • Add InvalidContractError by @yannham in #1824

Documentation

  • Add patterns to the syntax section of the manual by @yannham in #1832
  • Improve the description of identifier syntax in the manual by @vkleen in #1839
  • Add subsection on enum types in the manual by @yannham in #1836
  • Fix old CLI syntax in documentation by @cydparser in #1844
  • Add manual section on modular configurations by @yannham in #1841
  • Update/refresh examples using latest Nickel idioms by @yannham in #1849

Fixes

  • Check if stderr is a terminal for error messages by @vkleen in #1766
  • correctly drop Array::IntoIter by @Radvendii in #1773
  • LSP: vendor codespan, and fix character offset issues by @jneem in #1793
  • [Fix & Refactor] Row conflict error by @yannham in #1808
  • Fix unbound identifier when querying in REPL by @yannham in #1843
  • Fix --field not applying pending contracts by @yannham in #1778

Version 1.4

Nickel 1.4 is a maintenance release, at the exception of a breaking change (see below).

Breaking changes

  • The curried dot operator added in Nickel 1.3 was implemented the wrong way: the arguments were flipped, meaning that (.) foo bar was bar."%{foo}" instead of the expected foo."%{bar}". While the initial flipped implementation seems more useful for piping operations using the reverse application operator |>, it's inconsistent with all other curried operators, where (<operator>) is always defined as fun x y => x <operator> y. To ensure consistency, and because the initial behavior was an oversight and not a conscious design decision, we decided to change the definition of (.) to match the other curried operator by flipping its arguments.

    To fill the gap, Nickel 1.4 introduces std.record.get with the same definition as the (.) introduced in Nickel 1.3. To migrate from 1.3 to 1.4, you can either flip the arguments of the curried dot as a function (.) whenever possible, or you can just replace it with the new std.record.get.

    (implemented by @yannham in #1752)

Tooling

  • Search for imports in NICKEL_IMPORT_PATH by @jneem in #1716
  • Add a cli param --import-path to specify the search path by @jneem in #1721
  • LSP: Fix hover on assignments to subrecords by @jneem in #1725
  • Print something when nickel doc succeeds by @yannham in #1729
  • Add --error-format flag to serialize err diagnostics by @yannham in #1740
  • LSP: get record completion in arrays by @jneem in #1746

Core language

  • Support importing txt files as strings by @Quantum64 in #1734

Fixes

  • nickel format: don't fail silently on invalid input anymore by @yannham in #1749
  • Update Topiary dependencies to correctly handle (.) in nickel format by @yannham in #1753

Version 1.3

Version 1.3 includes several new optimizations following reports of long evaluation time for medium-sized codebase. The command-line interface (CLI) has been reworked to be more user-friendly, at the cost of breaking changes: see below. Finally, the LSP has seen continuous improvement as well.

Breaking changes

  • @vkleen improved the CLI UX in numerous ways in #1632
    • The file argument is now argument positional. That is, instead of running nickel export -f config.ncl, now use nickel export config.ncl instead.
    • Every command which can take a file argument can now take several of them. The program parsed from the files are then merged before applying the action. For example, the new nickel export config1.ncl config2.ncl config3.ncl is the equivalent of the previous: nickel export <<< '(import "config1.ncl") & (import "config2.ncl") & (import "config3.ncl")'
    • Evaluation is now an explicit subcommand, instead of being the default action. Instead of running nickel -f config.ncl to evaluate a file, use nickel eval config.ncl instead.
  • Not a breaking change per se, because the customize mode is experimental, but @yannham introduced a new syntax for customize mode in #1709. Instead of dynamically generating a CLI where arguments are field paths, the new customize mode CLI directly take assignments written in a Nickel-like syntax as positional arguments. For example, in 1.2, the command nickel eval -f confing.ncl -- \ --input.field1 '"Value"' --input.flag false \ --override output.bar 0 now becomes nickel eval config.ncl -- \ 'input.field1="Value"' input.flag=false \ --override output.bar=0

Fixes

  • Fix record.update by making record.insert act consistently by @yannham in #1669

Tooling

  • LSP:
    • implement type-based completion in the new completer by @jneem in #1577
    • Improve context completion by @jneem in #1584
    • Take the ancestor path into account when env-completing from uncles. by @jneem in #1661
    • Add goto support for pattern bindings by @jneem in #1665
    • Add cousin search to goto and hover by @jneem in #1670
    • Improve hover output for let patterns by @jneem in #1696
    • First prototype of contract evaluation by @jneem in #1672
  • LSP: a large refactoring work by @jneem to get rid of the old and hard-to-maintain code analysis implementation (#1623, #1629, #1658, #1663)
  • Honor nostdlib in customize mode as well by @vkleen in #1634
  • Add the list subcommand to the customize mode by @yannham in #1709
  • add %eval_nix% primop for Nix interop by @Radvendii in #1465 (requires to build with the corresponding experimental feature enabled)
  • Get rid of shallow_repr and print full terms in error messages by @yannham in #1676
  • Add suggestions to the error message when misspelling a record field by @yannham in #1710
  • Add a --field argument to subcommands to target a specific field whenever it makes sense by @yannham in #1712

Optimizations

  • Contract elision for static types by @yannham in #1671
  • Implement contract deduplication optimization by @yannham in #1631
  • Array contract deduplication by @yannham in #1674
  • Get rid of most generated variables by @yannham in #1679

Documentation

  • Fix invalid example code in doc of blame_with_message by @bgni in #1689
  • Fix doc, example code for pipe lacks prefix by @bgni in #1692
  • change nickel-nix to organist by @Radvendii in #1691

New Contributors

  • @bgni made their first contribution in #1689
  • @giorgiga made their first contribution in #1697

Version 1.2

Version 1.2 comes with several improvements on the LSP and other components of the Nickel tooling. The new customize mode of the CLI makes it possible to dynamically turn a configuration into a command-line interface, which you can interact with.

Several related long-standing issues and limitations when typechecking polymorphic functions are also finally fixed.

Core language

  • Make the lexer accept scientific notation by @vkleen in #1456
  • Improve polymorphism handling, bidirectional typechecking and fix unsound generalization by @yannham in #1372
  • Error on serializing very large numbers by @vkleen in #1470
  • Import YAML files containing multiple documents as arrays by @vkleen in #1497
  • Normalize line endings in string literals during parsing (Windows compatiblity) by @vkleen in #1562
  • Implement curried dot operator by @suimong in #1578
  • Statically merge partial record definitions by @vkleen in #1599
  • Disable recursive priorities by @yannham in #1600

Fixes

  • Various fixes to the pretty-printer by @vkleen (#1411, #1412, #1410, #1417)
  • Fix REPL panic on transitive imports by @Radvendii in #1474
  • Fix contract application order in let bindings and annotations by @vkleen in #1544
  • Fix cursor desyncing when using REPL by @deotimedev in #1546
  • Fix handling of relative imports @jneem in #1489
  • LSP Fix two issues with incomplete input by @jneem in #1550
  • Fix incomplete check of record row constraints by @yannham in #1558
  • Fix LSP regression on hovering by @yannham in #1583

Tooling

  • Full vscode extension by @szlend and @yannham in (#1405, #1413, #1416)
  • Invalidate importers in NLS (plus a couple other import-related issues) by @jneem in #1426
  • Improve nickel query (and :query in the REPL) interface by @yannham in #1447
  • Topiary integration (formatting) as nickel format by @vkleen in #1371
  • LSP: various improvements to completion by @jneem (#1450, #1473)
  • Improve doc extraction capabilities through nickel doc (evaluate terms before extracting documentation) by @vkleen in #1463
  • LSP Fix persistent diagnostics by @jneem in #1478
  • Restore --version for CLI, include git revision by @yannham in #1486
  • Disable the colors when stdout isn't a tty by @thufschmitt in #1508
  • Improve type variables name allocation when reporting type errors by @yannham in #1512
  • LSP formatting without calling the topiary binary by @vkleen in #1526
  • LSP Initial handling of incomplete input by @jneem in #1541
  • LSP resolve imports for incomplete inputs also by @jneem in #1542
  • Add multiline string support for VSCode autoclosing pairs by @deotimedev in #1553
  • Display icon for nickel file in vscode explorer by @suimong in #1556
  • New experimental customize mode CLI to interact with a configuration on the command line (see nickel help export) by @Radvendii and @yannham in #1475
  • (LSP) Import completion by @deotimedev in #1561
  • Get rid of uninformative generic notes for higher-order contract errors by @yannham in #1564
  • Render hover documentation as Markdown in LSP by @deotimedev in #1559
  • Exit with nonzero exit code on failure by @vkleen in #1576

Stdlib

  • add std.contract.Sequence by @Radvendii in #1452

Documentation

  • manual docs: fix mismatch between input command and its echo in the output by @foo-jin in #1421
  • Add two examples: imports and foreach pattern by @mipmip in #1387
  • Fix some leftover instances of the old enum syntax in the manual by @vkleen in #1548

Version 1.1

This version mostly includes bugfixes and stdlib improvement since 1.0.

IMPORTANT: The main crate nickel-lang has been split between nickel-lang-cli (the binary) and nickel-lang-core (the library). If you're using cargo to install Nickel, please uninstall the previous crate by running cargo uninstall nickel-lang, and from now one use cargo install nickel-lang-cli to install 1.1 and do further updates.

Stdlib

  • Encode more pre-conditions of stdlib functions as additional contracts, replacing dynamic type errors with nice contract errors by @yannham in #1358
  • ArraySliceFun incorrectly excluded the length of the array as the end index by @vkleen in #1396

Tooling

  • Improve the pretty printing of terms in the CLI and within error messages by @vkleen in #1262
  • Fix LSP panic when importing JSON by @yannham in #1382
  • Fix LSP hanging under certain conditions involving external imports by @yannham in #1390
  • Better error message when contract makes certain rows illegal by @Radvendii in #1323
  • Fix function params hovering in the LSP by @yannham in #1395
  • Fix LSP infinite loop on hovering on external imports by @yannham in #1397

Fixes

  • Fixes incorrect variable names in type mismatch errors by @vkleen in #1313
  • Improve error messages for polymorphic tail parametricity violations by @matthew-healy in #1320
  • Fix panic on duplicated top-level idents in record destructuring by @matthew-healy in #1324
  • Prevent panic when parsing record type with field path by @matthew-healy in #1325
  • Give a better error message when trying to query a non-record by @jneem in #1326
  • Fix error position by @jneem in #1333
  • Fix panic when interpolating fields in a record type by @jneem in #1332
  • Fix type annotations not generating contracts by @yannham in #1379
  • Fix typechecker looping by adding missing check for unifying equal type vars by @yannham in #1400
  • Fix panic on function contract failure (position of arrow functions not set) by @yannham in #1407

Version 1.0

This is the 1.0 release! The syntax and the semantics of the core language have been stabilized and shouldn't evolve as much in the future.

  • The core semantics of the language have been reworked and stabilized, in particular (but not limited to) merging, types and contracts with the implementation of RFC005
  • The stdlib has been augmented with many new functions
  • Parts of the syntax and some builtin symobls (types, stdlib functions, and so on) have been improved and made consistent
  • New features for the LSP, and in particular code completion

Breaking changes

  • Various functions from the stdlib have been renamed for better discoverability, and the stdlib got a lot of new additions. Please refer to the documentation of the stdlib.
  • String functions are now unicode-aware, and operate on the Unicode grapheme cluster abstraction instead of the character abstraction (string.length, string.is_match, etc.)
  • The switch keyword has been replaced by match, and can now be used as a standalone function (doesn't need to be applied right away)
  • The Num and Str builtin types have been renamed to Number and String by @yannham in #1164
  • The num and str stdlib modules have been renamed to number and string
  • The builtin.typeof function now returns 'Number, 'String, 'Function instead of respectively 'Num, 'Str, and 'Fun
  • The builtin.is_num, builtin.is_str and builtin.to_str functions have been renamed to is_number, is_string and to_string
  • The string.to_num and string.from_num functions have been renamed to to_number and from_number
  • All the stdlib modules array, string, record, etc. have been put under a std namespace. They must now be accessed as std.array, std.string and so on.
  • RFC005 was implemented, which changes the semantics of contract annotations and merging. See the RFC content for more details. Most notably, metadata annotation (default values, optional, documentation, etc.) can only appear next to a record field. Contract and type annotations can still appear anywhere. Documentation can still appear on let-bindings.
  • Use static dictionary types for record.fields and record.values by @matthew-healy in #1024
  • Make type annotations to not propagate through merging by @yannham in #1271
  • Change to dictionary contracts and introduction of a separate dictionary contract (in addition to dictionary type):
    • Fix dictionary contract semantics by @vkleen in #1141
    • Introduce dictionary contracts by @yannham in #1272
  • Stdlib string.Stringingable -> string.Stringable by @vkleen in #1180
  • Fix the type of array.elem by @yannham in #1223
  • Change the enum tag start delimiter from backtick to single-quote by @vkleen in #1279
  • import is now a statement, import "foo.ncl" arg1 arg2 requires parenthesis now: (import "foo.ncl") arg1 arg2, see #1293

Language features

  • Symbolic strings by @matthew-healy in #994
  • [RFC005] Lazy propagation by @yannham in #1086
  • Non-exported record fields by @vkleen in #1132
  • Enrich label custom reporting data by @yannham in #1152
  • Use type annotations in record patterns during typechecking by @matthew-healy in #1176
  • Use arbitrary precision rationals as the underlying representation of numbers by @yannham in #1182
  • Allow equal arrays to be merged in order to make merge idempotent by @yannham in #1229
  • Use deterministic hashtables, making runtime errors and typechecking error deterministic by @yannham in #1235

Stdlib

  • Add a %trace% primop (and std.trace function) by @vkleen in #1055
  • Add std.contract.Equal contract to the stdlib by @yannham in #1203

Tooling

  • Display meta information when providing completion in #966
  • LSP completion for import terms by @ebresafegaga in #993
  • Fix LSP server not giving completion when a non-contract meta-information is present in a declaration by @ebresafegaga in #991
  • Fix LSP not giving completion when an identifier is prefixed by a delimiting character by @ebresafegaga in #1043
  • Improve the "goto definition" feature of the LSP by making it work across multiple files by @ebresafegaga in #1029
  • Add --color option to the CLI by @matthew-healy in #1033
  • Make the "find references" feature of the LSP work across multiple files by @ebresafegaga in #1037
  • Add support for LSP completion using the surrounding context by @ebresafegaga (context completion) in #1057
  • Support completion for field names inside recursive records. by @ebresafegaga in #1088
  • Improve contract violation error reporting by @ebresafegaga in #1139
  • Add a JSON documentation export option by @vkleen in #1209
  • Add support for formatting capabilities to the LSP by @ebresafegaga in #1216

Fixes

  • Do not panic on type path mismatch by @yannham in #1028
  • Fix multiline string interpolation when preceded by a " character by @matthew-healy in #1023
  • Improve the performance of std.array.fold_left and std.array.fold_right by @vkleen in #1075
  • Correctly type identifiers introduced in destructuring patterns by @matthew-healy in #1099
  • Always include fields with a value in record.fields by @vkleen in #1225
  • Make partially applied boolean operators work by @vkleen in #1282

Version 0.3.1 (2022-12-15)

Fixes

  • Fix blame error when calling string.find by @euank in #988

Version 0.3 (2022-12-07)

Fixes

  • Fix polymorphic contracts unduly changing semantics by @ebresafegaga in #802
  • Fix typechecking and unification in presence of flat types (aka opaque types, aka contracts) by @yannham in #766
  • Fix polarity for polymorphic contract failure by @ebresafegaga in #831
  • Fix panic when a row mismatch occurs while unifying row tails by @matthew-healy in #847
  • Fix type to term conversion causing unbound type variables errors by @francois-caddet in #854
  • Fix bad lexing of enum tags by @matthew-healy in #874
  • Fix multiple recursive overriding by @yannham in #940

Language features

  • Optional fields by @yannham in #815
  • Numeral merge priorities by @yannham in #829
  • Recursive merge priorities (or push-priorities, or leafy priorities) by @yannham in #845
  • Change switch to match and make it a proper function by @yannham in #970

Stdlib

  • Statically type string.join by @matthew-healy in #946

Tooling

  • Add record completion in the LSP by @ebresafegaga in
  • Add completion for Nickel's stdlib in the LSP by @ebresafegaga in #918

Performances

  • Lazy array contracts by @fuzzypixelz in #809
  • Array slices by @fuzzypixelz in #776
  • String interning for identifiers by @Acaccia in #835

Version 0.2 (2022-07-29)

Breaking changes

  • Using a contract as part of a static type annotation will in most cases fail with an appropriate error message. This is a temporary limitation in order to fix previously unsound behavior in the typechecker. This restriction will likely be lifted in the upcoming 0.3.x release. For more details, see issues #701 and #724

Fixes

  • Fix unnecessarily restricted record contract for record.update
  • Fix wrong interpretation of long interpolation-like sequences %..%{ in strings
  • Fix panic when evaluating a switch in specific cases
  • Fix fields without definition being assigned to null, instead of just being marked as undefined

Language features

  • Merging null values together gives null, and merging empty lists together gives an empty list, instead of failing with error: non mergeable terms
  • Add recursive let-bindings (let rec)
  • Add type wildcards. Use _ in place of a type to let the typechecker fill the gap. Example: let foo : _ = array.all ((==) 2) [1,2,3]
  • Add builtin.to_str and string.from to convert generic values to a string
  • Re-introduce an official syntax for enum types

Tooling

  • Add the nickel pprint-ast command to pretty print a parsed program (mostly debugging purpose)
  • Add the nickel doc command to produce markdown documentation from the in-code doc metadata

Documentation

  • Fix various typos and remove use of deprecated syntax in the user manual

Version 0.1 (2022-03-10)

First release! The main focus has been the design of the Nickel language itself.

Language features

  • Gradual type system with row types, polymorphism and type inference
  • Contract system for data validation
  • Merge system for recursive records that supports one level of overriding
  • Metadata annotations (default values, documentation, etc.)
  • Unified syntax for terms, types and contracts (RFC002)
  • Record destructuring

Tooling

  • The main binary supports the following subcommands:

    • nickel query to show metadata and documentation of library functions, the field of a configuration, etc.
    • nickel export to serialize to JSON, YAML, or TOML
    • nickel repl to launch an REPL
    • nickel typecheck to do typechecking without evaluating

  • An LSP-server is included

Documentation

  • User manual sections on syntax, correctness (types and contracts), and merging
  • The standard library has been documented in-code (use nickel query/:query to retrieve it)

Known limitations

  • The roadmap for overriding and the merge system (RFC001) has not been implemented fully yet.

  • Performance has not been prioritized.

  • Due to the use of reference counting as a memory management strategy, mutually recursive record fields are currently leaking memory. This shouldn't be an issue in a standard workflow.

  • Standard library APIs and language features are subject to change. There is no backward compatibility guarantees for this version. In general, this release is meant for experimenting and getting user feedback, but isn't intended to be used in production.