Lexical 0.51 drops CommonJS from npm at the start of its clean-up before 1.0

Every package in the release now declares "type": "module", and its exports map has no require branch ^[5]. The release notes are direct about who escapes: "Bundlers are unaffected." ^[6] CommonJS code on Node.js 20.19 or later can still require() the packages. Anything older has to use await import(...) ^[7]. Metro bundles both builds unless the team adds the condition for Lexical's packages ^[9]. The CommonJS build still exists, but only in the source tree, kept for Meta's www ^[2].

A dependency loaded two ways is the harder case. The duplicate-copy failure applies to any app that loads yjs or @preact/signals-core through require() while Lexical loads them as ESM ^[8]. The project's advice is to load those as ESM too ^[8]. For a collaborative editor built on Yjs, that means checking how each shared dependency is loaded before taking the upgrade.

Serialization is where a build can pass and the output can still be wrong. Lexical's own export paths, including editorState.toJSON() and the clipboard export, start from the node map and are unaffected ^[11]. The risk is in custom code that keeps a node reference across a mutation, and the notes' fix is to call getLatest() on that reference first ^[11]. A third breaking change moves the Prism helpers out of @lexical/code into @lexical/code-prism ^[12].

Timing comes from the deprecation warning. This is a monthly release, and the next few are expected to focus on moving everything to the Extension and $config APIs and removing deprecated functionality ^[1][4]. "Expect anything @deprecated to be removed soon," the notes say ^[3]. At one release a month, that is a window of a few months ^[18]. The notes do not give a date for 1.0 or say which release removes which API.

LexicalComposer's entry is softer than that general warning. It is deprecated in favour of LexicalExtensionComposer and "expected to be removed in a future major" ^[13]. On a 0.x version line, the notes leave open whether that means 1.0 itself or something later. The project's case for moving anyway is functional: an editor built by LexicalComposer cannot reach any feature shipped as an extension ^[14].

For a team deciding this quarter, the trade-off is between a module change now and a bigger combined job later. A team that pins below 0.51 avoids the ESM work today. If the forecast removals land, it would then take the ESM switch and the deleted APIs in a single upgrade ^[19]. A team that moves now takes the module change while the deprecated APIs still exist, and can replace them one at a time before they go ^[13].

Compact JSON export is the one place where moving early is the wrong order. It is opt-in, and its raw objects are well under half the legacy size, though after gzip the two are a wash ^[15]. The smaller raw size matters for structured clones and worker messages, according to the notes ^[15]. The schemas behind it are still marked experimental, and default output is unchanged ^[17].