stack/packages/cli/tests/organization-imports.mjs

// Row 333 R2 (#1831): the files outside the organization package roots that its code imports, found
// by reading its imports. Rename r2 stage B: the roots are packages/organization/src, the
// implementation, and packages/business/src, the old-path shims of it; both must exist. A copy
// fixture carries these beside both roots, so a new import of
// a plug-in source file (row 333's model-quota.mjs) does not break the release tests the way it did
// when each fixture named kinds.mjs alone. A real copy ships all of packages/plugins (scripts/
// stack-copy.sh COPY_PATHS); a fixture models a smaller one and takes only what the loader reads.
//
// Vikunja #779: the helper reads imports with a regex, not a parser, so it stays strict about what
// it cannot read. Comments are blanked before matching, so an import in a comment is not one, and an
// import form inside a string or a regular expression is not one either. A dynamic import written
// with a space, `import ('./x.mjs')`, counts, and so does one after a spread, `{...import('./x.mjs')}`;
// a property call such as `x.import(...)` or `x?.import(...)` does not. `1..import(...)`, a property
// call on a number, reads as an import too, which errs toward refusal. These throw: a dynamic import
// whose argument is not a plain string, a dynamic import inside a template substitution (${...}), a
// file with an unterminated string, comment, template or regular expression, a target that is missing
// or not a regular file, and a target whose real path leaves the repository. A substitution with no
// dynamic import is blanked with the rest of the template, as is `${import.meta.url}`; the text of a
// template and a substitution that holds a template of its own are read to their ends.
// The scanner knows line and block comments, quoted strings, template literals, regular expression
// literals (by the usual guess from the previous token) and a hashbang. The guess is wrong for a
// regular expression after `)` or a keyword, as in `return /x/.test(y)`. After `++` or `--` a slash
// is division when the character before the pair ends an operand (`n++ / 2`) and starts a regular
// expression when it does not (`x = ++/re/.lastIndex`). That misreads one rare form as division: a
// prefix `++` or `--` applied to a regular expression's member right after `)`, `]`, a word or a
// line break, as in `if (c) ++/re/.lastIndex`.
import {readdirSync,readFileSync,realpathSync,statSync} from 'node:fs';
import {dirname,join,relative,resolve,sep} from 'node:path';

// `import` as a word that is not a property name. The one rule for both patterns below: a word or $
// before it, or a single dot (x.import, x?.import), makes it a property call; a dot that follows
// another dot is the end of a spread (...import), which evaluates the call, so it is not excluded.
const IMPORT_WORD=/(?<![\w$])(?<!(?<!\.)\.)import/.source;
const DYNAMIC=new RegExp(`${IMPORT_WORD}\\s*\\(`,'g');
const IMPORT=new RegExp([
 /(?:import|export)\s[^'"`;]*?\bfrom\s*['"](\.{1,2}\/[^'"]+)['"]/.source,
 /\bimport\s*['"](\.{1,2}\/[^'"]+)['"]/.source,
 `${DYNAMIC.source}\\s*['"](\\.{1,2}\\/[^'"]+)['"]\\s*\\)`,
].join('|'),'g');
const DYNAMIC_LITERAL=/^import\s*\(\s*(['"])[^'"\n]*\1\s*\)/;
// The plug-in loader's one computed import, allowed only in that file and only in this exact form. It loads
// packages/plugins/<kind>/<name>/index.mjs after checking that path, and a real copy ships all of
// packages/plugins, so a fixture that never loads a plug-in needs nothing more from it.
const LOADER_FILE='packages/plugins/src/loader.mjs',LOADER_IMPORT=/^import\(pathToFileURL\(expected\)\.href\)/;
const REGEX_AFTER='(,=:[!&|?{};+-*%<>~^';
const inert=(literals,index)=>literals.some(([from,to])=>index>=from&&index<to);
const sources=(dir)=>readdirSync(dir,{withFileTypes:true}).flatMap((e)=>e.isDirectory()?sources(join(dir,e.name)):e.name.endsWith('.mjs')?[join(dir,e.name)]:[]);

// The text with every comment blanked and every template literal emptied, and the ranges of the
// quoted strings and regular expressions left in it, as [from,to) offsets into the returned code.
// Throws on a string, comment, template or regular expression that never ends, and on a dynamic
// import inside a template substitution.
function scan(text){
 let code='',i=0,previous='',before='';
 const literals=[],open=[];
 const fail=(what)=>{throw new Error(`unterminated ${what}`);};
 const mark=(token)=>{before=previous;previous=token;};
 const keep=(end,last)=>{literals.push([code.length,code.length+end-i]);code+=text.slice(i,end);i=end;mark(last);};
 // Template text from i to the closing backtick, or to the ${ that opens a substitution, which is then read as code.
 const template=()=>{
  for(;i<text.length;i++){
   if(text[i]==='\\')i++;
   else if(text[i]==='`'){i++;mark('`');return;}
   else if(text[i]==='$'&&text[i+1]==='{'){i+=2;open.push({braces:0,start:code.length});mark('{');return;}
  }
  fail('template');
 };
 // A substitution is read as code to its closing brace, refused if it holds a dynamic import, then dropped.
 const close=({start})=>{
  for(const m of code.slice(start).matchAll(DYNAMIC))if(!inert(literals,start+m.index))throw new Error('unsupported import inside a template substitution');
  code=code.slice(0,start);
  while(literals.length&&literals[literals.length-1][0]>=start)literals.pop();
 };
 if(text.startsWith('#!')){const end=text.indexOf('\n');i=end<0?text.length:end;}
 while(i<text.length){
  const c=text[i],next=text[i+1];
  if(c==='/'&&next==='/'){const end=text.indexOf('\n',i);code+=' ';i=end<0?text.length:end;continue;}
  if(c==='/'&&next==='*'){const end=text.indexOf('*/',i+2);if(end<0)fail('comment');code+=' ';i=end+2;continue;}
  if(c==='"'||c==="'"){
   let j=i+1;
   for(;j<text.length&&text[j]!==c;j++){if(text[j]==='\\')j++;else if(text[j]==='\n')fail('string');}
   if(j>=text.length)fail('string');
   keep(j+1,c);continue;
  }
  if(c==='`'){code+='``';i++;template();continue;}
  if(c==='/'&&(previous===''||REGEX_AFTER.includes(previous))){
   let j=i+1,inClass=false;
   for(;j<text.length&&(inClass||text[j]!=='/');j++){
    if(text[j]==='\\')j++;else if(text[j]==='\n')fail('regular expression');
    else if(text[j]==='[')inClass=true;else if(text[j]===']')inClass=false;
   }
   if(j>=text.length)fail('regular expression');
   keep(j+1,'/');continue;
  }
  if(open.length&&(c==='{'||c==='}')){
   const top=open[open.length-1];
   if(c==='{')top.braces++;
   else if(top.braces===0){close(open.pop());i++;template();continue;}
   else top.braces--;
  }
  code+=c;i++;
  if(!/\s/.test(c)){
   const postfix=(c==='+'||c==='-')&&previous===c&&text[i-2]===c&&before!==''&&!REGEX_AFTER.includes(before);
   mark(postfix?c+c:c);
  }
 }
 if(open.length)fail('template');
 return {code,literals};
}

// The specifiers a source file imports by a relative path. A dynamic import of anything but a plain
// string throws, as does a file the scanner cannot read to its end.
function specifiers(file,text){
 let scanned;
 try{scanned=scan(text);}
 catch(error){throw new Error(`${file} cannot be read for imports: ${error.message}`);}
 const {code,literals}=scanned,live=(m)=>!inert(literals,m.index);
 for(const m of code.matchAll(DYNAMIC)){
  const at=code.slice(m.index,m.index+512);
  if(file===LOADER_FILE&&LOADER_IMPORT.test(at))continue;
  if(live(m)&&!DYNAMIC_LITERAL.test(at))throw new Error(`${file} has a dynamic import that is not a plain string literal`);
 }
 return [...code.matchAll(IMPORT)].filter(live).map((m)=>m[1]??m[2]??m[3]);
}

// The package roots a copy fixture carries whole. A missing root throws (readdirSync), so a tree
// with only one of them is refused rather than walked by half.
const ROOTS=[['packages','organization','src'],['packages','business','src']];

// Repo-relative paths, sorted: every file that a relative import from either root reaches outside
// both, following the imports of those files in turn. An import of a missing file, of anything
// but a regular file, or of a file whose real path is outside the repository throws, wherever the
// importing file sits, so an import of a missing file inside a root throws too.
export function organizationImportedFiles(repo){
 const owns=ROOTS.map((parts)=>join(repo,...parts)+sep),found=new Set(),checked=new Set(),todo=owns.flatMap((own)=>sources(own)),root=realpathSync(repo)+sep;
 for(let file;(file=todo.pop())!==undefined;){
  for(const spec of specifiers(relative(repo,file),readFileSync(file,'utf8'))){
   const target=resolve(dirname(file),spec),from=`${relative(repo,file)} imports ${spec}`;
   if(!checked.has(target)){
    let info,real;
    try{info=statSync(target);real=realpathSync(target);}
    catch(error){throw error.code==='ENOENT'||error.code==='ENOTDIR'?new Error(`${from}, which does not exist`):error;}
    if(!info.isFile())throw new Error(`${from}, which is not a regular file`);
    if(!real.startsWith(root))throw new Error(`${from}, which is outside the repository`);
    checked.add(target);
   }
   if(owns.some((own)=>target.startsWith(own))||found.has(target))continue;
   found.add(target);
   if(target.endsWith('.mjs'))todo.push(target);
  }
 }
 return [...found].map((f)=>relative(repo,f)).sort();
}