loader: sort files in scanning to force determinism

[nit.git] / src / loader.nit

# This file is part of NIT ( http://www.nitlanguage.org ).
#
# Copyright 2012 Jean Privat <jean@pryen.org>
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

# Loading of Nit source files
#
# The loader takes care of looking for module and projects in the file system, and the associated case of errors.
# The loading requires several steps:
#
# Identify: create an empty model entity associated to a name or a file path.
# Identification is used for instance when names are given in the command line.
# See `identify_module` and `identify_group`.
#
# Scan: visit directories and identify their contents.
# Scanning is done to enable the searching of modules in projects.
# See `scan_group` and `scan_full`.
#
# Parse: load the AST and associate it with the model entity.
# See `MModule::parse`.
#
# Import: means recursively load modules imported by a module.
# See `build_module_importation`.
#
# Load: means doing the full sequence: identify, parse and import.
# See `ModelBuilder::parse`, `ModelBuilder::parse_full`, `MModule::load` `ModelBuilder::load_module.
module loader

import modelbuilder_base
import ini

redef class ToolContext
        # Option --path
        var opt_path = new OptionArray("Set include path for loaders (may be used more than once)", "-I", "--path")

        # Option --only-metamodel
        var opt_only_metamodel = new OptionBool("Stop after meta-model processing", "--only-metamodel")

        # Option --only-parse
        var opt_only_parse = new OptionBool("Only proceed to parse step of loaders", "--only-parse")

        redef init
        do
                super
                option_context.add_option(opt_path, opt_only_parse, opt_only_metamodel)
        end
end

redef class ModelBuilder
        redef init
        do
                super

                # Setup the paths value
                paths.append(toolcontext.opt_path.value)

                var path_env = "NIT_PATH".environ
                if not path_env.is_empty then
                        paths.append(path_env.split_with(':'))
                end

                var nit_dir = toolcontext.nit_dir
                var libname = nit_dir/"lib"
                if libname.file_exists then paths.add(libname)
                libname = nit_dir/"contrib"
                if libname.file_exists then paths.add(libname)
        end

        # Load a bunch of modules.
        # `modules` can contains filenames or module names.
        # Imported modules are automatically loaded and modelized.
        # The result is the corresponding model elements.
        # Errors and warnings are printed with the toolcontext.
        #
        # Note: class and property model elements are not analysed.
        fun parse(modules: Sequence[String]): Array[MModule]
        do
                var time0 = get_time
                # Parse and recursively load
                self.toolcontext.info("*** PARSE ***", 1)
                var mmodules = new ArraySet[MModule]
                for a in modules do
                        var nmodule = self.load_module(a)
                        if nmodule == null then continue # Skip error
                        var mmodule = nmodule.mmodule
                        if mmodule == null then continue # skip error
                        mmodules.add mmodule
                end
                var time1 = get_time
                self.toolcontext.info("*** END PARSE: {time1-time0} ***", 2)

                self.toolcontext.check_errors

                if toolcontext.opt_only_parse.value then
                        self.toolcontext.info("*** ONLY PARSE...", 1)
                        exit(0)
                end

                return mmodules.to_a
        end

        # Identify a bunch of modules and groups.
        #
        # This does the same as `parse_full` but does only the identification (cf. `identify_module`)
        fun scan_full(names: Sequence[String]): Array[MModule]
        do
                var mmodules = new Array[MModule]
                for a in names do
                        # Case of a group (root or sub-directory)
                        var mgroup = self.identify_group(a)
                        if mgroup != null then
                                scan_group(mgroup)
                                for mg in mgroup.in_nesting.smallers do mmodules.add_all mg.mmodules
                                continue
                        end

                        # Case of a directory that is not a group
                        var stat = a.to_path.stat
                        if stat != null and stat.is_dir then
                                self.toolcontext.info("look in directory {a}", 2)
                                var fs = a.files
                                alpha_comparator.sort(fs)
                                # Try each entry as a group or a module
                                for f in fs do
                                        var af = a/f
                                        mgroup = identify_group(af)
                                        if mgroup != null then
                                                scan_group(mgroup)
                                                for mg in mgroup.in_nesting.smallers do mmodules.add_all mg.mmodules
                                                continue
                                        end
                                        var mmodule = identify_module(af)
                                        if mmodule != null then
                                                mmodules.add mmodule
                                        else
                                                self.toolcontext.info("ignore file {af}", 2)
                                        end
                                end
                                continue
                        end

                        var mmodule = identify_module(a)
                        if mmodule == null then
                                continue
                        end

                        mmodules.add mmodule
                end
                return mmodules
        end

        # Load a bunch of modules and groups.
        #
        # Each name can be:
        #
        # * a path to a module, a group or a directory of packages.
        # * a short name of a module or a group that are looked in the `paths` (-I)
        #
        # Then, for each entry, if it is:
        #
        # * a module, then is it parsed and returned.
        # * a group then recursively all its modules are parsed.
        # * a directory of packages then all the modules of all packages are parsed.
        # * else an error is displayed.
        #
        # See `parse` for details.
        fun parse_full(names: Sequence[String]): Array[MModule]
        do
                var time0 = get_time
                # Parse and recursively load
                self.toolcontext.info("*** PARSE ***", 1)
                var mmodules = new ArraySet[MModule]
                var scans = scan_full(names)
                for mmodule in scans do
                        var ast = mmodule.load(self)
                        if ast == null then continue # Skip error
                        mmodules.add mmodule
                end
                var time1 = get_time
                self.toolcontext.info("*** END PARSE: {time1-time0} ***", 2)

                self.toolcontext.check_errors

                if toolcontext.opt_only_parse.value then
                        self.toolcontext.info("*** ONLY PARSE...", 1)
                        exit(0)
                end

                return mmodules.to_a
        end

        # The list of directories to search for top level modules
        # The list is initially set with:
        #
        #   * the toolcontext --path option
        #   * the NIT_PATH environment variable
        #   * `toolcontext.nit_dir`
        # Path can be added (or removed) by the client
        var paths = new Array[String]

        # Like (and used by) `get_mmodule_by_name` but does not force the parsing of the MModule (cf. `identify_module`)
        fun search_mmodule_by_name(anode: nullable ANode, mgroup: nullable MGroup, name: String): nullable MModule
        do
                # First, look in groups
                var c = mgroup
                if c != null then
                        var r = c.mpackage.root
                        assert r != null
                        scan_group(r)
                        var res = r.mmodules_by_name(name)
                        if res.not_empty then return res.first
                end

                # Look at some known directories
                var lookpaths = self.paths

                # Look in the directory of the group package also (even if not explicitly in the path)
                if mgroup != null then
                        # path of the root group
                        var dirname = mgroup.mpackage.root.filepath
                        if dirname != null then
                                dirname = dirname.join_path("..").simplify_path
                                if not lookpaths.has(dirname) and dirname.file_exists then
                                        lookpaths = lookpaths.to_a
                                        lookpaths.add(dirname)
                                end
                        end
                end

                var candidate = search_module_in_paths(anode.hot_location, name, lookpaths)

                if candidate == null then
                        if mgroup != null then
                                error(anode, "Error: cannot find module `{name}` from `{mgroup.name}`. Tried: {lookpaths.join(", ")}.")
                        else
                                error(anode, "Error: cannot find module `{name}`. Tried: {lookpaths.join(", ")}.")
                        end
                        return null
                end
                return candidate
        end

        # Get a module by its short name; if required, the module is loaded, parsed and its hierarchies computed.
        # If `mgroup` is set, then the module search starts from it up to the top level (see `paths`);
        # if `mgroup` is null then the module is searched in the top level only.
        # If no module exists or there is a name conflict, then an error on `anode` is displayed and null is returned.
        fun get_mmodule_by_name(anode: nullable ANode, mgroup: nullable MGroup, name: String): nullable MModule
        do
                var mmodule = search_mmodule_by_name(anode, mgroup, name)
                if mmodule == null then return null # Forward error
                var ast = mmodule.load(self)
                if ast == null then return null # Forward error
                return mmodule
        end

        # Search a module `name` from path `lookpaths`.
        # If found, the module is returned.
        private fun search_module_in_paths(location: nullable Location, name: String, lookpaths: Collection[String]): nullable MModule
        do
                var res = new ArraySet[MModule]
                for dirname in lookpaths do
                        # Try a single module file
                        var mp = identify_module((dirname/"{name}.nit").simplify_path)
                        if mp != null then res.add mp
                        # Try the default module of a group
                        var g = identify_group((dirname/name).simplify_path)
                        if g != null then
                                scan_group(g)
                                res.add_all g.mmodules_by_name(name)
                        end
                end
                if res.is_empty then return null
                if res.length > 1 then
                        toolcontext.error(location, "Error: conflicting module files for `{name}`: `{[for x in res do x.filepath or else x.full_name].join("`, `")}`")
                end
                return res.first
        end

        # Search groups named `name` from paths `lookpaths`.
        private fun search_group_in_paths(name: String, lookpaths: Collection[String]): ArraySet[MGroup]
        do
                var res = new ArraySet[MGroup]
                for dirname in lookpaths do
                        # try a single group directory
                        var mg = identify_group(dirname/name)
                        if mg != null then
                                res.add mg
                        end
                end
                return res
        end

        # Cache for `identify_module` by relative and real paths
        private var identified_modules_by_path = new HashMap[String, nullable MModule]

        # All the currently identified modules.
        # See `identify_module`.
        #
        # An identified module exists in the model but might be not yet parsed (no AST), or not yet analysed (no importation).
        var identified_modules = new Array[MModule]

        # All the currently parsed modules.
        #
        # A parsed module exists in the model but might be not yet analysed (no importation).
        var parsed_modules = new Array[MModule]

        # Identify a source file and load the associated package and groups if required.
        #
        # This method does what the user expects when giving an argument to a Nit tool.
        #
        # * If `path` is an existing Nit source file (with the `.nit` extension),
        #   then the associated MModule is returned
        # * If `path` is a directory (with a `/`),
        #   then the MModule of its default module is returned (if any)
        # * If `path` is a simple identifier (eg. `digraph`),
        #   then the main module of the package `digraph` is searched in `paths` and returned.
        #
        # Silently return `null` if `path` does not exists or cannot be identified.
        #
        # On success, it returns a module that is possibly not yet parsed (no AST), or not yet analysed (no importation).
        # If the module was already identified, or loaded, it is returned.
        fun identify_module(path: String): nullable MModule
        do
                # special case for not a nit file
                if not path.has_suffix(".nit") then
                        # search dirless files in known -I paths
                        if not path.chars.has('/') then
                                var res = search_module_in_paths(null, path, self.paths)
                                if res != null then return res
                        end

                        # Found nothing? maybe it is a group...
                        var candidate = null
                        if path.file_exists then
                                var mgroup = identify_group(path)
                                if mgroup != null then
                                        var owner_path = mgroup.filepath.join_path(mgroup.name + ".nit")
                                        if owner_path.file_exists then candidate = owner_path
                                end
                        end

                        if candidate == null then
                                return null
                        end
                        path = candidate
                end

                # Does the file exists?
                if not path.file_exists then
                        return null
                end

                # Fast track, the path is already known
                if identified_modules_by_path.has_key(path) then return identified_modules_by_path[path]
                var rp = module_absolute_path(path)
                if identified_modules_by_path.has_key(rp) then return identified_modules_by_path[rp]

                var pn = path.basename(".nit")

                # Search for a group
                var mgrouppath = path.join_path("..").simplify_path
                var mgroup = identify_group(mgrouppath)

                if mgroup == null then
                        # singleton package
                        var mpackage = new MPackage(pn, model)
                        mgroup = new MGroup(pn, mpackage, null) # same name for the root group
                        mgroup.filepath = path
                        mpackage.root = mgroup
                        toolcontext.info("found singleton package `{pn}` at {path}", 2)

                        # Attach homonymous `ini` file to the package
                        var inipath = path.dirname / "{pn}.ini"
                        if inipath.file_exists then
                                var ini = new ConfigTree(inipath)
                                mpackage.ini = ini
                        end
                end

                var src = new SourceFile.from_string(path, "")
                var loc = new Location(src, 0, 0, 0, 0)
                var res = new MModule(model, mgroup, pn, loc)
                res.filepath = path

                identified_modules_by_path[rp] = res
                identified_modules_by_path[path] = res
                identified_modules.add(res)
                return res
        end

        # Groups by path
        private var mgroups = new HashMap[String, nullable MGroup]

        # Return the mgroup associated to a directory path.
        # If the directory is not a group null is returned.
        #
        # Note: `paths` is also used to look for mgroups
        fun identify_group(dirpath: String): nullable MGroup
        do
                var stat = dirpath.file_stat

                if stat == null then do
                        # search dirless directories in known -I paths
                        if dirpath.chars.has('/') then return null
                        for p in paths do
                                var try = p / dirpath
                                stat = try.file_stat
                                if stat != null then
                                        dirpath = try
                                        break label
                                end
                        end
                        return null
                end label

                # Filter out non-directories
                if not stat.is_dir then
                        return null
                end

                # Fast track, the path is already known
                var rdp = module_absolute_path(dirpath)
                if mgroups.has_key(rdp) then
                        return mgroups[rdp]
                end

                # By default, the name of the package or group is the base_name of the directory
                var pn = rdp.basename

                # Check `package.ini` that indicate a package
                var ini = null
                var parent = null
                var inipath = dirpath / "package.ini"
                if inipath.file_exists then
                        ini = new ConfigTree(inipath)
                end

                if ini == null then
                        # No ini, multiple course of action

                        # The root of the directory hierarchy in the file system.
                        if rdp == "/" then
                                mgroups[rdp] = null
                                return null
                        end

                        # Special stopper `packages.ini`
                        if (dirpath/"packages.ini").file_exists then
                                # dirpath cannot be a package since it is a package directory
                                mgroups[rdp] = null
                                return null
                        end

                        # check the parent directory (if it does not contain the stopper file)
                        var parentpath = dirpath.join_path("..").simplify_path
                        var stopper = parentpath / "packages.ini"
                        if not stopper.file_exists then
                                # Recursively get the parent group
                                parent = identify_group(parentpath)
                                if parent == null then
                                        # Parent is not a group, thus we are not a group either
                                        mgroups[rdp] = null
                                        return null
                                end
                        end
                end

                var mgroup
                if parent == null then
                        # no parent, thus new package
                        if ini != null then pn = ini["package.name"] or else pn
                        var mpackage = new MPackage(pn, model)
                        mgroup = new MGroup(pn, mpackage, null) # same name for the root group
                        mpackage.root = mgroup
                        toolcontext.info("found package `{mpackage}` at {dirpath}", 2)
                        mpackage.ini = ini
                else
                        mgroup = new MGroup(pn, parent.mpackage, parent)
                        toolcontext.info("found sub group `{mgroup.full_name}` at {dirpath}", 2)
                end

                # search documentation
                # in src first so the documentation of the package code can be distinct for the documentation of the package usage
                var readme = dirpath.join_path("README.md")
                if not readme.file_exists then readme = dirpath.join_path("README")
                if readme.file_exists then
                        var mdoc = load_markdown(readme)
                        mgroup.mdoc = mdoc
                        mdoc.original_mentity = mgroup
                end

                mgroup.filepath = dirpath
                mgroups[rdp] = mgroup
                return mgroup
        end

        # Load a markdown file as a documentation object
        fun load_markdown(filepath: String): MDoc
        do
                var s = new FileReader.open(filepath)
                var lines = new Array[String]
                var line_starts = new Array[Int]
                var len = 1
                while not s.eof do
                        var line = s.read_line
                        lines.add(line)
                        line_starts.add(len)
                        len += line.length + 1
                end
                s.close
                var source = new SourceFile.from_string(filepath, lines.join("\n"))
                source.line_starts.add_all line_starts
                var mdoc = new MDoc(new Location(source, 1, lines.length, 0, 0))
                mdoc.content.add_all(lines)
                return mdoc
        end

        # Force the identification of all MModule of the group and sub-groups in the file system.
        #
        # When a group is scanned, its sub-groups hierarchy is filled (see `MGroup::in_nesting`)
        # and the potential modules (and nested modules) are identified (see `MGroup::modules`).
        #
        # Basically, this recursively call `identify_group` and `identify_module` on each directory entry.
        #
        # No-op if the group was already scanned (see `MGroup::scanned`).
        fun scan_group(mgroup: MGroup) do
                if mgroup.scanned then return
                mgroup.scanned = true
                var p = mgroup.filepath
                # a virtual group has nothing to scan
                if p == null then return
                var files = p.files
                alpha_comparator.sort(files)
                for f in files do
                        var fp = p/f
                        var g = identify_group(fp)
                        # Recursively scan for groups of the same package
                        if g == null then
                                identify_module(fp)
                        else if g.mpackage == mgroup.mpackage then
                                scan_group(g)
                        end
                end
        end

        # Transform relative paths (starting with '../') into absolute paths
        private fun module_absolute_path(path: String): String do
                return path.realpath
        end

        # Try to load a module AST using a path.
        # Display an error if there is a problem (IO / lexer / parser) and return null
        #
        # The AST is loaded as is total independence of the model and its entities.
        #
        # AST are not cached or reused thus a new AST is returned on success.
        fun load_module_ast(filename: String): nullable AModule
        do
                if not filename.has_suffix(".nit") then
                        self.toolcontext.error(null, "Error: file `{filename}` is not a valid nit module.")
                        return null
                end
                if not filename.file_exists then
                        self.toolcontext.error(null, "Error: file `{filename}` not found.")
                        return null
                end

                self.toolcontext.info("load module {filename}", 2)

                # Load the file
                var file = new FileReader.open(filename)
                var lexer = new Lexer(new SourceFile(filename, file))
                var parser = new Parser(lexer)
                var tree = parser.parse
                file.close

                # Handle lexer and parser error
                var nmodule = tree.n_base
                if nmodule == null then
                        var neof = tree.n_eof
                        assert neof isa AError
                        error(neof, neof.message)
                        return null
                end

                return nmodule
        end

        # Remove Nit source files from a list of arguments.
        #
        # Items of `args` that can be loaded as a nit file will be removed from `args` and returned.
        fun filter_nit_source(args: Array[String]): Array[String]
        do
                var keep = new Array[String]
                var res = new Array[String]
                for a in args do
                        var l = identify_module(a)
                        if l == null then
                                keep.add a
                        else
                                res.add a
                        end
                end
                args.clear
                args.add_all(keep)
                return res
        end

        # Try to load a module using a path.
        # Display an error if there is a problem (IO / lexer / parser) and return null.
        # Note: usually, you do not need this method, use `get_mmodule_by_name` instead.
        #
        # The MModule is located, created, parsed and the importation is performed.
        fun load_module(filename: String): nullable AModule
        do
                # Look for the module
                var mmodule = identify_module(filename)
                if mmodule == null then
                        if filename.file_exists then
                                toolcontext.error(null, "Error: `{filename}` is not a Nit source file.")
                        else
                                toolcontext.error(null, "Error: cannot find module `{filename}`.")
                        end
                        return null
                end

                # Load it
                return mmodule.load(self)
        end

        # Injection of a new module without source.
        # Used by the interpreter.
        fun load_rt_module(parent: nullable MModule, nmodule: AModule, mod_name: String): nullable AModule
        do
                # Create the module

                var mgroup = null
                if parent != null then mgroup = parent.mgroup
                var mmodule = new MModule(model, mgroup, mod_name, nmodule.location)
                nmodule.mmodule = mmodule
                nmodules.add(nmodule)
                self.mmodule2nmodule[mmodule] = nmodule

                if parent!= null then
                        var imported_modules = new Array[MModule]
                        imported_modules.add(parent)
                        mmodule.set_visibility_for(parent, intrude_visibility)
                        mmodule.set_imported_mmodules(imported_modules)
                else
                        build_module_importation(nmodule)
                end

                return nmodule
        end

        # Visit the AST and create the `MModule` object
        private fun build_a_mmodule(mgroup: nullable MGroup, nmodule: AModule)
        do
                var mmodule = nmodule.mmodule
                assert mmodule != null

                # Check the module name
                var decl = nmodule.n_moduledecl
                if decl != null then
                        var decl_name = decl.n_name.n_id.text
                        if decl_name != mmodule.name then
                                error(decl.n_name, "Error: module name mismatch; declared {decl_name} file named {mmodule.name}.")
                        end
                end

                # Check for conflicting module names in the package
                if mgroup != null then
                        var others = model.get_mmodules_by_name(mmodule.name)
                        if others != null then for other in others do
                                if other != mmodule and mmodule2nmodule.has_key(mmodule) and other.mgroup!= null and other.mgroup.mpackage == mgroup.mpackage then
                                        var node: ANode
                                        if decl == null then node = nmodule else node = decl.n_name
                                        error(node, "Error: a module named `{other.full_name}` already exists at {other.location}.")
                                        break
                                end
                        end
                end

                nmodules.add(nmodule)
                self.mmodule2nmodule[mmodule] = nmodule

                var source = nmodule.location.file
                if source != null then
                        assert source.mmodule == null
                        source.mmodule = mmodule
                end

                if decl != null then
                        # Extract documentation
                        var ndoc = decl.n_doc
                        if ndoc != null then
                                var mdoc = ndoc.to_mdoc
                                mmodule.mdoc = mdoc
                                mdoc.original_mentity = mmodule
                        else
                                advice(decl, "missing-doc", "Documentation warning: Undocumented module `{mmodule}`")
                        end
                        # Is the module a test suite?
                        mmodule.is_test_suite = not decl.get_annotations("test_suite").is_empty
                end
        end

        # Resolve the module identification for a given `AModuleName`.
        #
        # This method handles qualified names as used in `AModuleName`.
        fun seach_module_by_amodule_name(n_name: AModuleName, mgroup: nullable MGroup): nullable MModule
        do
                var mod_name = n_name.n_id.text

                # If a quad is given, we ignore the starting group (go from path)
                if n_name.n_quad != null then mgroup = null

                # If name not qualified, just search the name
                if n_name.n_path.is_empty then
                        # Fast search if no n_path
                        return search_mmodule_by_name(n_name, mgroup, mod_name)
                end

                # If qualified and in a group
                if mgroup != null then
                        # First search in the package
                        var r = mgroup.mpackage.root
                        assert r != null
                        scan_group(r)
                        # Get all modules with the final name
                        var res = r.mmodules_by_name(mod_name)
                        # Filter out the name that does not match the qualifiers
                        res = [for x in res do if match_amodulename(n_name, x) then x]
                        if res.not_empty then
                                if res.length > 1 then
                                        error(n_name, "Error: conflicting module files for `{mod_name}`: `{[for x in res do x.filepath or else x.full_name].join("`, `")}`")
                                end
                                return res.first
                        end
                end

                # If no module yet, then assume that the first element of the path
                # Is to be searched in the path.
                var root_name = n_name.n_path.first.text
                var roots = search_group_in_paths(root_name, paths)
                if roots.is_empty then
                        error(n_name, "Error: cannot find `{root_name}`. Tried: {paths.join(", ")}.")
                        return null
                end

                var res = new ArraySet[MModule]
                for r in roots do
                        # Then, for each root, collect modules that matches the qualifiers
                        scan_group(r)
                        var root_res = r.mmodules_by_name(mod_name)
                        for x in root_res do if match_amodulename(n_name, x) then res.add x
                end
                if res.not_empty then
                        if res.length > 1 then
                                error(n_name, "Error: conflicting module files for `{mod_name}`: `{[for x in res do x.filepath or else x.full_name].join("`, `")}`")
                        end
                        return res.first
                end
                # If still nothing, just call a basic search that will fail and will produce an error message
                error(n_name, "Error: cannot find module `{mod_name}` from `{root_name}`. Tried: {paths.join(", ")}.")
                return null
        end

        # Is elements of `n_name` correspond to the group nesting of `m`?
        #
        # Basically it check that `bar::foo` matches `bar/foo.nit` and `bar/baz/foo.nit`
        # but not `baz/foo.nit` nor `foo/bar.nit`
        #
        # Is used by `seach_module_by_amodule_name` to validate qualified names.
        private fun match_amodulename(n_name: AModuleName, m: MModule): Bool
        do
                var g: nullable MGroup = m.mgroup
                for grp in n_name.n_path.reverse_iterator do
                        while g != null and grp.text != g.name do
                                g = g.parent
                        end
                end
                return g != null
        end

        # Analyze the module importation and fill the module_importation_hierarchy
        #
        # If the importation was already done (`nmodule.is_importation_done`), this method does a no-op.
        #
        # REQUIRE `nmodule.mmodule != null`
        # ENSURE `nmodule.is_importation_done`
        fun build_module_importation(nmodule: AModule)
        do
                if nmodule.is_importation_done then return
                nmodule.is_importation_done = true
                var mmodule = nmodule.mmodule.as(not null)
                var stdimport = true
                var imported_modules = new Array[MModule]
                for aimport in nmodule.n_imports do
                        # Do not imports conditional
                        var atconditionals = aimport.get_annotations("conditional")
                        if atconditionals.not_empty then continue

                        stdimport = false
                        if not aimport isa AStdImport then
                                continue
                        end

                        # Load the imported module
                        var sup = seach_module_by_amodule_name(aimport.n_name, mmodule.mgroup)
                        if sup == null then
                                mmodule.is_broken = true
                                nmodule.mmodule = null # invalidate the module
                                continue # Skip error
                        end
                        var ast = sup.load(self)
                        if ast == null then
                                mmodule.is_broken = true
                                nmodule.mmodule = null # invalidate the module
                                continue # Skip error
                        end

                        aimport.mmodule = sup
                        imported_modules.add(sup)
                        var mvisibility = aimport.n_visibility.mvisibility
                        if mvisibility == protected_visibility then
                                mmodule.is_broken = true
                                error(aimport.n_visibility, "Error: only properties can be protected.")
                                mmodule.is_broken = true
                                nmodule.mmodule = null # invalidate the module
                                return
                        end
                        if sup == mmodule then
                                error(aimport.n_name, "Error: dependency loop in module {mmodule}.")
                                mmodule.is_broken = true
                                nmodule.mmodule = null # invalidate the module
                        end
                        if sup.in_importation < mmodule then
                                error(aimport.n_name, "Error: dependency loop between modules {mmodule} and {sup}.")
                                mmodule.is_broken = true
                                nmodule.mmodule = null # invalidate the module
                                return
                        end
                        mmodule.set_visibility_for(sup, mvisibility)
                end
                if stdimport then
                        var mod_name = "core"
                        var sup = self.get_mmodule_by_name(nmodule, null, mod_name)
                        if sup == null then
                                mmodule.is_broken = true
                                nmodule.mmodule = null # invalidate the module
                        else # Skip error
                                imported_modules.add(sup)
                                mmodule.set_visibility_for(sup, public_visibility)
                        end
                end

                # Declare conditional importation
                for aimport in nmodule.n_imports do
                        if not aimport isa AStdImport then continue
                        var atconditionals = aimport.get_annotations("conditional")
                        if atconditionals.is_empty then continue

                        var suppath = seach_module_by_amodule_name(aimport.n_name, mmodule.mgroup)
                        if suppath == null then continue # skip error

                        for atconditional in atconditionals do
                                var nargs = atconditional.n_args
                                if nargs.is_empty then
                                        error(atconditional, "Syntax Error: `conditional` expects module identifiers as arguments.")
                                        continue
                                end

                                # The rule
                                var rule = new Array[MModule]

                                # First element is the goal, thus
                                rule.add suppath

                                # Second element is the first condition, that is to be a client of the current module
                                rule.add mmodule

                                # Other condition are to be also a client of each modules indicated as arguments of the annotation
                                for narg in nargs do
                                        var id = narg.as_id
                                        if id == null then
                                                error(narg, "Syntax Error: `conditional` expects module identifier as arguments.")
                                                continue
                                        end

                                        var mp = search_mmodule_by_name(narg, mmodule.mgroup, id)
                                        if mp == null then continue

                                        rule.add mp
                                end

                                conditional_importations.add rule
                        end
                end

                mmodule.set_imported_mmodules(imported_modules)

                apply_conditional_importations(mmodule)

                self.toolcontext.info("{mmodule} imports {mmodule.in_importation.direct_greaters.join(", ")}", 3)

                # Force `core` to be public if imported
                for sup in mmodule.in_importation.greaters do
                        if sup.name == "core" then
                                mmodule.set_visibility_for(sup, public_visibility)
                        end
                end

                # TODO: Correctly check for useless importation
                # It is even doable?
                var directs = mmodule.in_importation.direct_greaters
                for nim in nmodule.n_imports do
                        if not nim isa AStdImport then continue
                        var im = nim.mmodule
                        if im == null then continue
                        if directs.has(im) then continue
                        # This generates so much noise that it is simpler to just comment it
                        #warning(nim, "Warning: possible useless importation of {im}")
                end
        end

        # Global list of conditional importation rules.
        #
        # Each rule is a "Horn clause"-like sequence of modules.
        # It means that the first module is the module to automatically import.
        # The remaining modules are the conditions of the rule.
        #
        # Rules are declared by `build_module_importation` and are applied by `apply_conditional_importations`
        # (and `build_module_importation` that calls it).
        #
        # TODO (when the loader will be rewritten): use a better representation and move up rules in the model.
        private var conditional_importations = new Array[SequenceRead[MModule]]

        # Extends the current importations according to imported rules about conditional importation
        fun apply_conditional_importations(mmodule: MModule)
        do
                # Because a conditional importation may cause additional conditional importation, use a fixed point
                # The rules are checked naively because we assume that it does not worth to be optimized
                var check_conditional_importations = true
                while check_conditional_importations do
                        check_conditional_importations = false

                        for ci in conditional_importations do
                                # Check conditions
                                for i in [1..ci.length[ do
                                        var m = ci[i]
                                        # Is imported?
                                        if not mmodule.in_importation.greaters.has(m) then continue label
                                end
                                # Still here? It means that all conditions modules are loaded and imported

                                # Identify the module to automatically import
                                var sup = ci.first
                                var ast = sup.load(self)
                                if ast == null then continue

                                # Do nothing if already imported
                                if mmodule.in_importation.greaters.has(sup) then continue label

                                # Import it
                                self.toolcontext.info("{mmodule} conditionally imports {sup}", 3)
                                # TODO visibility rules (currently always public)
                                mmodule.set_visibility_for(sup, public_visibility)
                                # TODO linearization rules (currently added at the end in the order of the rules)
                                mmodule.set_imported_mmodules([sup])

                                # Prepare to reapply the rules
                                check_conditional_importations = true
                        end label
                end
        end

        # All the loaded modules
        var nmodules = new Array[AModule]

        # Register the nmodule associated to each mmodule
        #
        # Public clients need to use `mmodule2node` to access stuff.
        private var mmodule2nmodule = new HashMap[MModule, AModule]

        # Retrieve the associated AST node of a mmodule.
        # This method is used to associate model entity with syntactic entities.
        #
        # If the module is not associated with a node, returns null.
        fun mmodule2node(mmodule: MModule): nullable AModule
        do
                return mmodule2nmodule.get_or_null(mmodule)
        end
end

redef class MModule
        # The path of the module source
        var filepath: nullable String = null

        # Force the parsing of the module using `modelbuilder`.
        #
        # If the module was already parsed, the existing ASI is returned.
        # Else the source file is loaded, and parsed and some
        #
        # The importation is not done by this
        #
        # REQUIRE: `filepath != null`
        # ENSURE: `modelbuilder.parsed_modules.has(self)`
        fun parse(modelbuilder: ModelBuilder): nullable AModule
        do
                # Already known and loaded? then return it
                var nmodule = modelbuilder.mmodule2nmodule.get_or_null(self)
                if nmodule != null then return nmodule

                var filepath = self.filepath
                assert filepath != null
                # Load it manually
                nmodule = modelbuilder.load_module_ast(filepath)
                if nmodule == null then return null # forward error

                # build the mmodule
                nmodule.mmodule = self
                modelbuilder.build_a_mmodule(mgroup, nmodule)

                modelbuilder.parsed_modules.add self
                return nmodule
        end

        # Parse and process importation of a given MModule.
        #
        # Basically chains `parse` and `build_module_importation`.
        fun load(modelbuilder: ModelBuilder): nullable AModule
        do
                var nmodule = parse(modelbuilder)
                if nmodule == null then return null

                modelbuilder.build_module_importation(nmodule)
                return nmodule
        end
end

redef class MPackage
        # The associated `.ini` file, if any
        #
        # The `ini` file is given as is and might contain invalid or missing information.
        #
        # Some packages, like stand-alone packages or virtual packages have no `ini` file associated.
        var ini: nullable ConfigTree = null
end

redef class MGroup
        # Is the group interesting for a final user?
        #
        # Groups are mandatory in the model but for simple packages they are not
        # always interesting.
        #
        # A interesting group has, at least, one of the following true:
        #
        # * it has 2 modules or more
        # * it has a subgroup
        # * it has a documentation
        fun is_interesting: Bool
        do
                return mmodules.length > 1 or
                        not in_nesting.direct_smallers.is_empty or
                        mdoc != null or
                        (mmodules.length == 1 and default_mmodule == null)
        end

        # Are files and directories in self scanned?
        #
        # See `ModelBuilder::scan_group`.
        var scanned = false

        # Return the modules in self and subgroups named `name`.
        #
        # If `self` is not scanned (see `ModelBuilder::scan_group`) the
        # results might be partial.
        fun mmodules_by_name(name: String): Array[MModule]
        do
                var res = new Array[MModule]
                for g in in_nesting.smallers do
                        for mp in g.mmodules do
                                if mp.name == name then
                                        res.add mp
                                end
                        end
                end
                return res
        end
end

redef class SourceFile
        # Associated mmodule, once created
        var mmodule: nullable MModule = null
end

redef class AStdImport
        # The imported module once determined
        var mmodule: nullable MModule = null
end

redef class AModule
        # The associated MModule once build by a `ModelBuilder`
        var mmodule: nullable MModule = null

        # Flag that indicate if the importation is already completed
        var is_importation_done: Bool = false
end