Summary

  • In Sketchpad, an instance referred to a master picture while carrying its own location, size and rotation; it was not an independently editable duplicate.
  • Master pictures could contain instances, so display expansion composed transformations through a recursive hierarchy. Changing the master changed dependent appearances without erasing their structural arrangement.
  • That propagation demonstrated a software dependency, not manufacture or physical correctness. Those claims require separate evidence about constraints, outputs, materials, tests and the exact revision used.

Draw one hexagon. Mark its corners as attachment points. Place seven instances into a larger pattern, then make an instance of that group and repeat. Now replace the original hexagon with a semicircle. In Ivan Sutherland’s description of Sketchpad, the whole pattern changes at once, yet the hierarchy that assembled it remains.

That demonstration is often celebrated as an ancestor of computer-aided design, graphical interfaces and object-oriented ideas. Its narrower technical lesson is just as important. The screen did not contain a pile of copied hexagons. It contained references to a shared definition, each interpreted through a local transformation. The result looked repetitive because the underlying dependency was repetitive.

A reference, not a duplicate

Sutherland’s January 1963 MIT dissertation calls a reusable subpicture an “instance” of a “master picture.” The instance is compact because it does not reproduce every line, arc and nested symbol in storage. It points back to the picture that defines its visible internal structure.

An instance still has a value of its own. Four numerical components locate its centre and encode its displayed size and rotation. Two instances of the same master may therefore appear at different coordinates, at different scales and at different angles. Shared definition does not mean identical placement.

The distinction becomes sharper in the dissertation’s account of copying. An instance retains its identity as one drawing part and has fixed internal structure. A copy of a definition picture loses that unit identity: its constituent parts can be deleted or changed individually. Sketchpad could even “dismember” an instance into a copy, peeling away one layer of shared identity. The vocabulary describes different data relationships, not two casual names for the same operation.

This matters whenever a later system presents a button labelled duplicate, clone, component, symbol, template or instance. The label alone does not establish whether the result remains linked to an authority, takes a snapshot, inherits future changes or becomes locally editable. Sketchpad’s record makes the question inspectable.

Transformations travel through the hierarchy

A master picture could itself contain instances. Displaying an outer instance therefore required Sketchpad to follow the reference, encounter inner instances, and continue until it reached picture parts with no further instances. At each level it relocated the parts into the correct position, size and rotation. For an instance inside another instance, the new transformation was joined to the transformation already in force.

That is more than storage economy. It gives the drawing a dependency path. A small symbol can become part of a group, that group can become part of a larger assembly, and one edit can be visible through every path that reaches the master. Sutherland identified recursive expansion—subpictures inside subpictures to as many levels as desired—as one of Sketchpad’s general functions.

The dissertation also records limits instead of concealing them. Mirror images were not represented by the instance value and had to be handled separately. Circular references were prevented from expanding without limit; Sutherland noted that a simple termination rule was difficult because a recursive instance could grow rather than shrink. “Recursive” did not mean unconstrained magic.

Appearance can propagate while meaning does not

Sketchpad stored topology and constraints alongside appearance. If a polygon vertex moved, adjacent lines moved with it. Attachment points kept wires or other drawing elements connected to symbols. Conditions could require lines to be equal, parallel, horizontal or perpendicular. These relationships made the drawing behave differently from a trail of carbon on paper.

But an instance did not inherit proof that the pictured thing would work in the world. Sutherland wrote that richer design conditions—materials, cost and other limitations—could be added to the system’s vocabulary so that a computer might help produce a sound design. The conditional wording marks the boundary. The implemented drawing and constraint machinery was not a general certificate of engineering validity.

Even within the drawing, constraint satisfaction was not guaranteed for an impossible collection of conditions. A successful display showed what Sketchpad had represented and solved. It did not show that a bridge had been built, that a circuit had been electrically verified, that a component met tolerances or that every relevant load case was present.

That separation is easy to lose in modern workflows. A rendered family of components may all update from one source and still be based on the wrong source. A parametric model may be internally consistent and still omit a manufacturing constraint. A generated drawing may look authoritative while referring to a master version that has not passed review.

Credit follows the evidence

Sketchpad was Sutherland’s dissertation and implementation, but the record does not portray solitary work without infrastructure or advice. Claude E. Shannon was the thesis supervisor. The acknowledgements credit Shannon and Marvin Minsky for help and advice, Steven A. Coons and Douglas T. Ross for time and ideas, and MIT Lincoln Laboratory for extensive support.

Wesley A. Clark and Jack L. Mitchell made TX-2 available and helped provide special equipment. Sutherland later recalled that Clark had designed TX-2 with interactive work in mind and immediately approved the proposed drawing investigation. In a Computer History Museum oral history, Sutherland described Coons as encouraging, said he talked extensively with Ross about structures and n-component items, and remembered Shannon pushing the system beyond straight lines by asking for circles.

Those facts set useful boundaries. Clark’s machine and access mattered; Shannon’s supervision and intervention mattered; Ross’s structural thinking and Coons’s encouragement mattered. None of those roles should be inflated into authorship of Sketchpad, and Sutherland’s authorship should not erase the institutional conditions that made the experiment possible.

The durable control surface

The master-instance mechanism converts repetition into governance. Whoever controls the master can alter every dependent appearance. Whoever defines whether an element is an instance or a copy controls whether local divergence is possible. Whoever records transforms and attachment paths controls whether an output can later be reconstructed.

A modern dependency receipt inspired by this boundary would preserve the master’s stable identity and version; the complete instance path; every composed position, size and rotation transform; relevant attachments and constraints; the change that triggered regeneration; the affected-output set; and the validation state of each output. A fabrication or physical-test record would join to that exact revision rather than to the family name alone.

Sketchpad did not provide that contemporary receipt. It provided the conceptual reason to demand one. Shared definition is powerful because it makes change travel. The same fact makes an unrecorded master change a wide blast radius.

Sources