Macro Debugger: Inspecting Macro Expansion

The macro-debugger collection contains two tools: a stepper for macro expansion and a standalone syntax browser. The macro stepper shows the programmer the expansion of a program as a sequence of rewriting steps, using the syntax browser to display the individual terms. The syntax browser uses colors and a properties panel to show the term’s syntax properties, such as lexical binding information and source location.

1 Macro Stepper

The repl is implemented by installing a custom evaluation handler that chains to the original handler to do evaluation.

2 Macro Expansion Tools

This module provides expand-like procedures that allow the user to specify macros whose expansions should be hidden.

Warning: because of limitations in the way macro expansion is selectively hidden, the resulting syntax may not evaluate to the same result as the original syntax.

3 Macro Stepper API for Macros

Macros can explicitly send information to a listening macro stepper by using the procedures in this module.

By default, syntax objects in remarks have the transformer’s mark applied (using syntax-local-introduce) so that their appearance in the macro stepper matches their appearance after the transformer returns. Unmarking is suppressed if unmark? is #f.

(define-syntax (mymac stx)

(syntax-case stx ()

[(_ x y)

(emit-remark "I got some arguments!"

#'x

"and"

#'y)

#'(list 'x 'y)]))

(mymac 37 (+ 1 2))

(Run the fragment above in the macro stepper.)

The before and after terms are marked with syntax-local-introduce so they appear in the macro stepper like they would if the step were truly generated from a local expansion.

The id argument acts as the step’s “macro” for the purposes of macro hiding.

4 Macro Stepper Text Interface

5 Syntax Browser

6 Using the Macro Stepper

6.1 Navigation

The stepper presents expansion as a linear sequence of rewriting process, and it gives the user controls to step forward or backwards as well as to jump to the beginning or end of the expansion process.

If the macro stepper is showing multiple expansions, then it also provides “Previous term” and “Next term” buttons to go up and down in the list of expansions. Horizontal lines delimit the current expansion from the others.

6.2 Macro Hiding

Macro hiding lets one see how expansion would look if certain macros were actually primitive syntactic forms. The macro stepper skips over the expansion of the macros you designate as opaque, but it still shows the expansion of their subterms.

The bottom panel of the macro stepper controls the macro hiding policy. The user changes the policy by selecting an identifier in the syntax browser pane and then clicking one of “Hide module”, “Hide macro”, or “Show macro”. The new rule appears in the policy display, and the user may later remove it using the "Delete" button.

The stepper also offers coarser-grained options that can hide collections of modules at once. These options have lower precedence than the rules above.

Macro hiding, even with no macros marked opaque, also hides certain other kinds of steps: internal defines are not rewritten to letrecs, begin forms are not spliced into module or block bodies, etc.

7 Using the Syntax Browser

7.1 Selection

The selection is indicated by bold text.

The user can click on any part of a subterm to select it. To select a parenthesized subterm, click on either of the parentheses. The selected syntax is bolded. Since one syntax object may occur inside of multiple other syntax objects, clicking on one occurrence will cause all occurrences to be bolded.

The syntax browser displays information about the selected syntax object in the properties panel on the right, when that panel is shown. The selected syntax also determines the highlighting done by the secondary partitioning (see below).

7.2 Primary Partition

The primary partition is indicated by foreground color.

The primary partitioning always assigns two syntax subterms the same color if they have the same marks. In the absence of unhygienic macros, this means that subterms with the same foreground color were either present in the original pre-expansion syntax or generated by the same macro transformation step.

Syntax colored in black always corresponds to unmarked syntax. Such syntax may be original, or it may be produced by the expansion of a nonhygienic macro.

Note: even terms that have the same marks might not be bound-identifier=? to each other, because they might occur in different environments.

7.3 Secondary Partitioning

The user may select a secondary partitioning through the Syntax menu. This partitioning applies only to identifiers. When the user selects an identifier, all terms in the same equivalence class as the selected term are highlighted in yellow.

7.4 Properties

When the properties pane is shown, it displays properties of the selected syntax object. The properties pane has two tabbed pages:

• Term:

  If the selection is an identifier, shows the binding information associated with the syntax object. For more information, see identifier-binding, etc.

• Syntax Object:

  Displays source location information and other properties (see syntax-property) carried by the syntax object.

7.5 Interpreting Syntax

The binding information of a syntax object may not be the same as the binding structure of the program it represents. The binding structure of a program is only determined after macro expansion is complete.

8 Finding Useless requires

The “Check Requires” utility can be run as a raco subcommand. For example (from racket root directory):

  raco check-requires racket/collects/syntax/*.rkt

  raco check-requires -kbu openssl

Each argument is interpreted as a file path if it exists; otherwise, it is interpreted as a module path. See check-requires for a description of the output format, known limitations in the script’s recommendations, etc.

Modules required for-label are not analyzed.

KEEP req-module at req-phase

The require of module req-module at phase req-phase must be kept because bindings defined within it are used.

If show-uses? is true, the dependencies of module-to-analyze on req-module are enumerated, one per line, in the following format:

exp-name at use-phase (mode ...) [RENAMED TO ref-name]

Indicates an export named exp-name is used at phase use-phase (not necessarily the phase it was provided at, if req-phase is non-zero).

The modes indicate what kind(s) of dependencies were observed: used as a reference, appeared in a syntax template (quote-syntax), etc.

If the RENAMED TO clause is present, it indicates that the binding is renamed on import into the module, and ref-name gives the local name used (exp-name is the name under which req-module provides the binding).

BYPASS req-module at req-phase

The require is used, but only for bindings that could be more directly obtained via one or more other modules. For example, a use of racket might be bypassed in favor of racket/base, racket/match, and racket/contract, etc.

A list of replacement requires is given, one per line, in the following format:

TO repl-module at repl-phase [WITH RENAMING]

Add a require of repl-module at phase repl-phase. If show-uses? is true, then following each TO line is an enumeration of the dependencies that would be satisfied by repl-module in the same format as described under KEEP below.

If the WITH RENAMING clause is present, it indicates that at least one of the replacement modules provides a binding under a different name from the one used locally in the module. Either the references should be changed or rename-in should be used with the replacement modules as necessary.

Bypass recommendations are restricted by the following rules:

• repl-module must not involve crossing a new private directory from req-module

• repl-module is never a built-in (“#%”) module

• req-module must not be in the “no-bypass” whitelist

DROP req-module at req-phase

The require appears to be unused, and it can probably be dropped entirely.

Due to limitations in its implementation strategy, check-requires occasionally suggests dropping or bypassing a module that should not be dropped or bypassed. The following are typical reasons for such bad suggestions:

• The module’s invocation has side-effects. For example, the module body may update a shared table or perform I/O, or it might transitively require a module that does. (Consider adding the module to the whitelist.)

• Bindings from the module are used in identifier comparisons by a macro, such as appearing in the macro’s “literals list.” In such cases, a macro should annotate its expansion with the 'disappeared-use property containing the identifier(s) compared with its literals; however, most casually-written macros do not do so. On the other hand, macros and their literal identifiers are typically provided by the same module, so this problem is somewhat uncommon.

9 Showing Module Dependencies

The “Show Dependencies” utility can be run as a raco subcommand. For example (from racket root directory):

  raco show-dependencies -bc racket/collects/openssl/main.rkt

  raco show-dependencies -c --exclude racket openssl

Each argument is interpreted as a file path if it exists; otherwise it is interpreted as a module path. See show-dependencies for a description of the output format.

Dependencies are printed, one per line, in the following format:

dep-module [<- (direct-dependent ...)]

Indicates that dep-module is transitively required by one or more root modules. If show-context? is true, then the direct-dependents are shown; they are the modules reachable from (and including) the root modules that directly require dep-module.

The dependencies are trimmed by removing any module reachable from (or equal to) a module in exclude as well as any module reachable from (but not equal to) a module in exclude-deps.

10 Macro Profiler

The Macro Profiler shows what macros contribute most to the expanded code size of programs. Use the Macro Profiler when your program has compiled files that are larger than expected. (The Macro Profiler does not report expansion time, but expansion time is generally proportional to code size.)

  raco macro-profiler module-path ...

The Macro Profiler works by expanding the files using the Macro Debugger and recording the difference in term sizes for each macro expansion step. The size of a term is computed by counting its pairs, atoms, etc.

Consider the following partial macro expansion:

(m (n x)) ⇒ (o (p y) (n x)) ⇒ (f (p y) (n x) z) ⇒ (f y (n x) z)

The direct cost of m is 6—the size of the new term (p y) plus one for the additional pair to include it in the o arguments. Likewise, the direct cost of o is 2. The direct cost of p is -4, because the macro’s result is smaller than its use.

The total cost of a macro consists of its direct cost plus the costs of any macros in the code introduced by m, but not including the costs from macro arguments like (n x). So the total cost of m is 6 + 2 - 4 = 4, because the o and p terms were introduced by m. In contrast, the total cost of o is just 2, the same as the direct cost.