Import from dhtmlxGantt#

Load a dhtmlxGantt project into Planning Engine and get the same task dates the gantt auto-scheduler would give you.

Planning Engine's scheduling step is a port of the dhtmlxGantt v2 auto-scheduling engine, and its JSON input uses gantt's field names (start_date, duration, parent, constraint_type, auto_scheduling). Gantt's link type values "0" to "3", task type: "project", and assignment task_id / resource_id are accepted as they are. Fields the engine doesn't know, such as text or open, are ignored in task, link, resource and assignment records. Most of the import is a straight copy. The steps below cover the places where a straight copy gives different dates or fails validation.

Prerequisites#

Your gantt data as plain JS objects, for example from gantt.serialize().
Each task's effective calendar, as gantt resolves it.
The module loaded as shown in Quick Start.

Steps#

Set the scheduling config to gantt's model. Use calendar_mode: "task" for task calendars only, or "task-inherited" if gantt has inherit_calendar: true. Copy gantt's work_time into use_calendars explicitly. If you leave use_calendars out, Planning Engine schedules on calendars:
```
const config = {
  calendar_mode: "task",
  use_calendars: ganttWorkTime, // gantt config.work_time
};
```
Set root_id only when gantt's virtual root isn't 0. The other auto-scheduling options (apply_constraints, gap_behavior, move_projects, schedule_from_end, project_start, use_progress) keep their gantt names.
Add one CalendarData entry per calendar your tasks use, and set each task's calendar_id to its effective calendar. gantt picks that calendar in its adapter: the task's own, else its resource's, else a calendar merged from all its resources, else the parent's. Send the result, not the rules. A merged resource calendar goes in once, under its synthetic id. Don't repeat gantt's precedence or merging in your import code.
Send dates as YYYY-MM-DD HH:mm strings (YYYY-MM-DD and a T separator also work). If gantt's date_format gives another format, such as day first, convert the dates first. Otherwise the call fails with a validation error, "invalid date".
Fix up each task:
- A summary (type: "project") takes both start_date and end_date, or neither. Drop its duration. Strict validation rejects a summary that has start_date and duration but no end_date.
- A parent equal to root_id means "no parent". You can copy it as is.
- A leaf task needs two of start_date, end_date and duration. If you send all three, they must agree.
- If your gantt event handlers cancel auto-scheduling for a task, set auto_scheduling: false on it. The planner never moves it.
Copy links as they are, type and lag included. Leave lag_unit out, because gantt measures lag in config.duration_unit.
Convert each assignment to a slice of its task's work, and drop gantt's mode:
- default and fixedDuration: rename delay to offset and keep duration. Leave both out to book the whole task. A delay you forget to rename is ignored as an unknown field, so the assignment silently books from the task start.
- fixedDates: set offset and duration to the worked time its dates cover, measured from the task start.
A slice ends at the task end. gantt's fixedDuration can run past it.
Pass the data to solve and write the changed fields from result.updates back into gantt. The updates include summaries, which the gantt JS adapter leaves to gantt core. Apply those too, or your summary spans go stale.

The full example below imports a summary with two tasks, a milestone, two links, a part-time calendar and one default-mode assignment:

const { JsonProject } = require("./pkg-node/plen_wasm.js");

function unwrap(text) {
  const response = JSON.parse(text);
  if (!response.ok) {
    const { kind, code, message, path } = response.error;
    throw new Error(`${kind} ${code ?? ""}: ${message} (at ${path})`);
  }
  return response.result;
}

// gantt.serialize() output, with each task's effective calendar already resolved.
const gantt = {
  data: [
    { id: 1, text: "Release", type: "project", parent: 0, start_date: "2026-03-02 00:00", end_date: "2026-03-07 00:00", duration: 5, calendar_id: "global" },
    { id: 2, text: "Build", parent: 1, start_date: "2026-03-02 00:00", duration: 3, calendar_id: "global" },
    { id: 3, text: "Test", parent: 1, start_date: "2026-03-02 00:00", duration: 4, calendar_id: "part-time" },
    { id: 4, text: "Ship", type: "milestone", parent: 0, start_date: "2026-03-02 00:00", duration: 0, calendar_id: "global" },
  ],
  links: [
    { id: 1, source: 2, target: 3, type: "0", lag: 0 },
    { id: 2, source: 1, target: 4, type: "0", lag: 0 },
  ],
  resources: [{ id: "dev" }],
  assignments: [{ id: 1, task_id: 3, resource_id: "dev", value: 1, mode: "default", delay: 1, duration: 2 }],
};

function task(t) {
  if (t.type !== "project") return t;
  // A summary takes both dates or neither, never a duration.
  const { duration, ...summary } = t;
  if (!summary.end_date) delete summary.start_date;
  return summary;
}

function assignment({ mode, delay, ...a }) {
  // default / fixedDuration: gantt's delay is the slice offset.
  return { ...a, offset: delay };
}

const data = {
  config: {
    calendar_mode: "task", // gantt's calendar model
    use_calendars: true,   // gantt config.work_time
  },
  calendars: [
    { id: "global" },
    { id: "part-time", worktime: { days: [false, true, true, true, true, false, false] } }, // Mon-Thu
  ],
  tasks: gantt.data.map(task),
  links: gantt.links,
  resources: gantt.resources,
  assignments: gantt.assignments.map(assignment),
};

const project = new JsonProject("{}");
try {
  const result = unwrap(project.solve(JSON.stringify(data), "{}"));
  for (const m of result.updates) {
    // Only changed fields are set. Summaries are included: write them back too.
    if (m.start_date) console.log(`${m.id} start_date = ${m.start_date}`);
    if (m.end_date) console.log(`${m.id} end_date = ${m.end_date}`);
  }
} finally {
  project.free();
}

Keep gantt's triggers on your side#

Planning Engine has no UI triggers and no events. Your code decides when to schedule.

Call apply and then solve(undefined, ...) wherever gantt would run auto-scheduling: after the user adds a link, drags a task, or saves the lightbox. See Editing a project.
gantt.autoSchedule(id, inclusive) is the subtree solve option. See Reschedule a branch.
Change sets can't add or edit calendars. If an assignment change gives a task a new merged calendar, send the whole project again with solve(dataJson, ...).
gantt core rewrites descendant dates when inherited calendars or effort change. Forward those date and duration edits as ordinary TaskUpdate changes. Planning Engine doesn't recreate gantt's effort rules.

Differences to expect#

Strict validation rejects input that gantt coerces: duplicate ids, a dated constraint with no constraint_date, a leaf task with fewer than two of start, end and duration, a dateless summary with no dated children. To get gantt's coercions with a report of each fix, see Accept messy input.
week, month and year duration units and calendars with customWeeks are rejected. A gantt merged calendar can contain customWeeks, so such projects can't be imported yet.
Dates are naive local date-times with no timezone and no DST. gantt's JS engine uses the host time zone, so its arithmetic follows DST. Convert dates to naive local time before you build the data.
result.notes lists the links, tasks and assignments the engine dropped. gantt drops them silently.
Three cases give different dates on purpose. With project_constraint: true, children inherit the parent's constraint, and gantt's engine ignores it. With use_progress, a completed child of a moving summary stays in place, while gantt drags it along. With descendant_links: true in block mode, gantt fails with a cycle error and Planning Engine schedules the project. See Links, constraints and summaries.
Summary-to-summary links in block mode, combined with leaf links between the same summaries, can give different dates. On very large generated graphs the call can trap with a RuntimeError. See Error handling.

Result#

Run the example with node. It prints the changed fields:

3 start_date = 2026-03-05 00:00
3 end_date = 2026-03-12 00:00
1 end_date = 2026-03-12 00:00
4 start_date = 2026-03-12 00:00
4 end_date = 2026-03-12 00:00

Task 3 starts when task 2 finishes and works Monday to Thursday only, so it ends a week later. The summary 1 grows to cover it. Task 2 didn't move, so it isn't listed. To get every date on the first load, see Load all task dates.