Developer reference

Internal global variables

These are set during execution of Revise's __init__ function.


Returns true if we watch files rather than their containing directory. FreeBSD and NFS-mounted systems should watch files, otherwise we prefer to watch directories.


Returns true if we should poll the filesystem for changes to the files that define loaded code. It is preferable to avoid polling, instead relying on operating system notifications via FileWatching.watch_file. However, NFS-mounted filesystems (and perhaps others) do not support file-watching, so for code stored on such filesystems you should turn polling on.

See the documentation for the JULIA_REVISE_POLL environment variable.


Returns true if files directly included from the REPL should be tracked. The default is false. See the documentation regarding the JULIA_REVISE_INCLUDE environment variable to customize it.


Constant specifying full path to julia top-level source directory. This should be reliable even for local builds, cross-builds, and binary installs.


Full path to the running Julia's cache of source code defining Base.


Julia's top-level directory when Julia was built, as recorded by the entries in Base._included_files.


Internal state management


pkgdatas is the core information that tracks the relationship between source code and julia objects, and allows re-evaluation of code in the proper module scope. It is a dictionary indexed by PkgId: pkgdatas[id] returns a value of type Revise.PkgData.


Global variable, watched_files[dirname] returns the collection of files in dirname that we're monitoring for changes. The returned value has type Revise.WatchList.

This variable allows us to watch directories rather than files, reducing the burden on the OS.


Global variable, revision_queue holds (pkgdata,filename) pairs that we need to revise, meaning that these files have changed since we last processed a revision. This list gets populated by callbacks that watch directories for updates.


Global variable; default PkgId used for files which do not belong to any package, but still have to be watched because user callbacks have been registered for them.


Global variable, maps (pkgdata, filename) pairs that errored upon last revision to (exception, backtrace).


Global variable, included_files gets populated by callbacks we register with include. It's used to track non-precompiled packages and, optionally, user scripts (see docs on JULIA_REVISE_INCLUDE).


The following are specific to user callbacks (see Revise.add_callback) and the implementation of entr:


This Condition is used to notify entr that one of the watched files has changed.


Global variable, user_callbacks_queue holds key values for which the file has changed but the user hooks have not yet been called.


Global variable, maps files (identified by their absolute path) to the set of callback keys registered for them.




A RelocatableExpr wraps an Expr to ensure that comparisons between RelocatableExprs ignore line numbering information. This allows one to detect that two expressions are the same no matter where they appear in a file.


For a particular source file, the corresponding ModuleExprsSigs is a mapping mod=>exprs=>sigs of the expressions exprs found in mod and the signatures sigs that arise from them. Specifically, if mes is a ModuleExprsSigs, then mes[mod][ex] is a list of signatures that result from evaluating ex in mod. It is possible that this returns nothing, which can mean either that ex does not define any methods or that the signatures have not yet been cached.

The first mod key is guaranteed to be the module into which this file was included.

To create a ModuleExprsSigs from a source file, see Revise.parse_source.

FileInfo(mexs::ModuleExprsSigs, cachefile="")

Structure to hold the per-module expressions found when parsing a single file. mexs holds the Revise.ModuleExprsSigs for the file.

Optionally, a FileInfo can also record the path to a cache file holding the original source code. This is applicable only for precompiled modules and Base. (This cache file is distinct from the original source file that might be edited by the developer, and it will always hold the state of the code when the package was precompiled or Julia's Base was built.) When a cache is available, mexs will be empty until the file gets edited: the original source code gets parsed only when a revision needs to be made.

Source cache files greatly reduce the overhead of using Revise.

PkgData(id, path, fileinfos::Dict{String,FileInfo})

A structure holding the data required to handle a particular package. path is the top-level directory defining the package, and fileinfos holds the Revise.FileInfo for each file defining the package.

For the PkgData associated with Main (e.g., for files loaded with includet), the corresponding path entry will be empty.


A struct for holding files that live inside a directory. Some platforms (OSX) have trouble watching too many files. So we watch parent directories, and keep track of which files in them should be tracked.


  • timestamp: mtime of last update
  • trackedfiles: Set of filenames, generally expressed as a relative path
thunk = TaskThunk(f, args)

To facilitate precompilation and reduce latency, we avoid creation of anonymous thunks. thunk can be used as an argument in schedule(Task(thunk)).

ReviseEvalException(loc::String, exc::Exception, stacktrace=nothing)

Provide additional location information about exc.

When running via the interpreter, the backtraces point to interpreter code rather than the original culprit. This makes it possible to use loc to provide information about the frame backtrace, and even to supply a fake backtrace.

If stacktrace is supplied it must be a Vector{Any} containing (::StackFrame, n) pairs where n is the recursion count (typically 1).


Create a portable summary of a method. In particular, a MethodSummary can be saved to a JLD2 file.


Function reference

Functions called during initialization of Revise


Wait for the REPL to complete its initialization, and then call Revise.steal_repl_backend. This is necessary because code registered with atreplinit runs before the REPL is initialized, and there is no corresponding way to register code to run after it is complete.

steal_repl_backend(backend = Base.active_repl_backend)

Replace the REPL's normal backend with one that calls revise before executing any REPL input.


Functions called when you load a new package


Start watching a package for changes to the files that define it. This function gets called via a callback registered with Base.require, at the completion of module-loading by using or import.


This function gets called by watch_package and runs when a package is first loaded. Its job is to organize the files and expressions defining the module so that later we can detect and process revisions.


Monitoring for changes

These functions get called on each directory or file that you monitor for revisions. These block execution until the file(s) are updated, so you should only call them from within an @async block. They work recursively: once an update has been detected and execution resumes, they schedule a revision (see Revise.revision_queue) and then call themselves on the same directory or file to wait for the next set of changes.

The following functions support user callbacks, and are used in the implementation of entr but can be used more broadly:

key = Revise.add_callback(f, files, modules=nothing; key=gensym())

Add a user-specified callback, to be executed during the first run of revise() after a file in files or a module in modules is changed on the file system. If all is set to true, also execute the callback whenever any file already monitored by Revise changes. In an interactive session like the REPL, Juno or Jupyter, this means the callback executes immediately before executing a new command / cell.

You can use the return value key to remove the callback later (Revise.remove_callback) or to update it using another call to Revise.add_callback with key=key.


Remove a callback previously installed by a call to Revise.add_callback(...). See its docstring for details.


Evaluating changes (revising) and computing diffs

revise is the primary entry point for implementing changes. Additionally,

Revise.revise_file_now(pkgdata::PkgData, file)

Process revisions to file. This parses file and computes an expression-level diff between the current state of the file and its most recently evaluated state. It then deletes any removed methods and re-evaluates any changed expressions. Note that generally it is better to use revise as it properly handles methods that move from one file to another.

id must be a key in Revise.pkgdatas, and file a key in Revise.pkgdatas[id].fileinfos.


Caching the definition of methods

success = get_def(method::Method)

As needed, load the source file necessary for extracting the code defining method. The source-file defining method must be tracked. If it is in Base, this will execute track(Base) if necessary.

This is a callback function used by CodeTracking.jl's definition.


Parsing source code

mexs = parse_source(filename::AbstractString, mod::Module)

Parse the source filename, returning a ModuleExprsSigs mexs. mod is the "parent" module for the file (i.e., the one that included the file); if filename defines more module(s) then these will all have separate entries in mexs.

If parsing filename fails, nothing is returned.

parse_source!(mexs::ModuleExprsSigs, filename, mod::Module)

Top-level parsing of filename as included into module mod. Successfully-parsed expressions will be added to mexs. Returns mexs if parsing finished successfully, otherwise nothing is returned.

See also Revise.parse_source.

success = parse_source!(mod_exprs_sigs::ModuleExprsSigs, src::AbstractString, filename::AbstractString, mod::Module)

Parse a string src obtained by reading file as a single string. pos is the 1-based byte offset from which to begin parsing src.

See also Revise.parse_source.


Lowered source code

Much of the "brains" of Revise comes from doing analysis on lowered code. This part of the package is not as well documented.

isrequired, evalassign = minimal_evaluation!([predicate,] methodinfo, src::Core.CodeInfo, mode::Symbol)

Mark required statements in src: isrequired[i] is true if src.code[i] should be evaluated. Statements are analyzed by isreq, haseval = predicate(stmt), and predicate defaults to Revise.is_method_or_eval. haseval is true if the statement came from @eval or eval(...) call. Since the contents of such expression are difficult to analyze, it is generally safest to execute all such evals.

methods_by_execution!(recurse=JuliaInterpreter.Compiled(), methodinfo, docexprs, mod::Module, ex::Expr;
                      mode=:eval, disablebp=true, skip_include=mode!==:eval, always_rethrow=false)

Evaluate or analyze ex in the context of mod. Depending on the setting of mode (see the Extended help), it supports full evaluation or just the minimal evaluation needed to extract method signatures. recurse controls JuliaInterpreter's evaluation of any non-intercepted statement; likely choices are JuliaInterpreter.Compiled() or JuliaInterpreter.finish_and_return!. methodinfo is a cache for storing information about any method definitions (see CodeTrackingMethodInfo). docexprs is a cache for storing documentation expressions; obtain an empty one with Revise.DocExprs().

Extended help

The action depends on mode:

  • :eval evaluates the expression in mod, similar to Core.eval(mod, ex) except that methodinfo and docexprs will be populated with information about any signatures or docstrings. This mode is used to implement includet.
  • :sigs analyzes ex and extracts signatures of methods and docstrings (specifically, statements flagged by Revise.minimal_evaluation!), but does not evaluate ex in the traditional sense. It will selectively execute statements needed to form the signatures of defined methods. It will also expand any @evaled expressions, since these might contain method definitions.
  • :evalmeth analyzes ex and extracts signatures and docstrings like :sigs, but takes the additional step of evaluating any :method statements.
  • :evalassign acts similarly to :evalmeth, and also evaluates assignment statements.

When selectively evaluating an expression, Revise will incorporate required dependencies, even for minimal-evaluation modes like :sigs. For example, the method definition

max_values(T::Union{map(X -> Type{X}, Base.BitIntegerSmall_types)...}) = 1 << (8*sizeof(T))

found in base/abstractset.jl requires that it create the anonymous function in order to compute the signature.

The other keyword arguments are more straightforward:

  • disablebp controls whether JuliaInterpreter's breakpoints are disabled before stepping through the code. They are restored on exit.
  • skip_include prevents execution of include statements, instead inserting them into methodinfo's cache. This defaults to true unless mode is :eval.
  • always_rethrow, if true, causes an error to be thrown if evaluating ex triggered an error. If false, the error is logged with @error. InterruptExceptions are always rethrown. This is primarily useful for debugging.

Create a cache for storing information about method definitions. Adding signatures to such an object inserts them into CodeTracking.method_info, which maps signature Tuple-types to (lnn::LineNumberNode, ex::Expr) pairs. Because method signatures are unique within a module, this is the foundation for identifying methods in a manner independent of source-code location.

It also has the following fields:

  • exprstack: used when descending into @eval statements (via push_expr and pop_expr!) ex (used in creating the CodeTrackingMethodInfo object) is the first entry in the stack.
  • allsigs: a list of all method signatures defined by a given expression
  • deps: list of top-level named objects (Symbols and GlobalRefs) that method definitions in this block depend on. For example, if Sys.iswindows() f() = 1 else f() = 2 end would store Sys.iswindows here.
  • includes: a list of module=>filename for any include statements encountered while the expression was parsed.

Modules and paths

parentfile, included_files = modulefiles(mod::Module)

Return the parentfile in which mod was defined, as well as a list of any other files that were included to define mod. If this operation is unsuccessful, (nothing, nothing) is returned.

All files are returned as absolute paths.


Handling errors


Truncate a list of instruction pointers, as obtained from backtrace() or catch_backtrace(), at the first "top-level" call (e.g., as executed from the REPL prompt) or the first entry corresponding to a method in Revise or its dependencies.

This is used to make stacktraces obtained with Revise more similar to those obtained without Revise, while retaining one entry to reveal Revise's involvement.


In current releases of Julia, hitting Ctrl-C from the REPL can stop tasks running in the background. This risks stopping Revise's ability to watch for changes in files and directories. Revise has a work-around for this problem.

success = throwto_repl(e::Exception)

Try throwing e from the REPL's backend task. Returns true if the necessary conditions were met and the throw can be expected to succeed. The throw is generated from another task, so a yield will need to occur before it happens.


Git integration

Revise.git_source(file::AbstractString, reference)

Read the source-text for file from a git commit reference. The reference may be a string, Symbol, or LibGit2.Tree.


Revise.git_source("/path/to/myfile.jl", "HEAD")
Revise.git_source("/path/to/myfile.jl", :abcd1234)  # by commit SHA
repo, repo_path = git_repo(path::AbstractString)

Return the repo::LibGit2.GitRepo containing the file or directory path. path does not necessarily need to be the top-level directory of the repository. Also returns the repo_path of the top-level directory for the repository.


Distributed computing


Define methods on worker p that Revise needs in order to perform revisions on p. Revise itself does not need to be running on p.