Nesc-internals

Driver Scripts

• ncc: driver script for TinyOS, understands TinyOS platforms, etc. Behaves like gcc with extra options. Main task is to generate and execute a nescc command

• nescc: driver script for nesC, understands how to compile a nesC application. Supports cross-compilation. Behaves like gcc with extra options. Main task is to generate and execute a gcc command with the appropriate magic (see the tdspecs file) to invoke nesc-compile on .nc files.

• mig, nescc-mig: TinyOS and non-TinyOS scripts to invoke the message interface generator. End up calling nescc with the "right" options.

• ncg, nescc-ncg: TinyOS and non-TinyOS scripts to invoke the constant generator. End up calling nescc with the "right" options.

Internal Script

• nesc-compile: Invoked by gcc when passed a .nc file. Must generate a .s or .o file depending on options received. Essentially, invokes nesc1 to transform nesC application into a C file, then invokes gcc to compile this C file.

In all these scripts, gcc stands for the target platform's gcc, though, except for the invocation from nesc-compile, this isn't typically crucial.

nesc1

Overview: takes a "root" .nc file and generates a C file representing the whole nesC application. Also has a few other paths, to generate information for mig, TinyOS 1.x doc info (legacy only, the 2.x docs use the XML output), and a C-with-nesC extensions to C path (new). Accepts gcc options and nescc options, as filtered by nescc/nesc-compile.

Implementation is based on a C frontend derived from gcc 2.8.1, with various additions of code from later versions of gcc. Ideally it should support the current gcc C syntax, but that ends up being more an on-demand thing -- when someone complains that some header file doesn't work, I add support for whatever new C feature is involved. Essentially what's missing is some of the newer ISO C99 features, and all #pragma's. As of 1.3.x, the preprocessor is integrated in nesc1 (nesc1 used to invoke the target gcc to perform preprocessing of nesC files).

Directory Structure

• top-level: the usual configuration stuff, READMEs, etc. Bootstrap is a script that runs automake, autoconf, etc to generate configure

• doc: man pages, reference manual, and miscellaneous documentation

• include, libiberty, libcpp: integrated C preprocessor, imported from gcc in early 2008. Very minor (a few lines) changes (sorry, there should be a nice version number and diff, but it isn't very hard to track down exactly which snapshot I used, and hence recompute the diff).

• tests: old stuff, ignore.

• nregress: regression tests, execute ./runtest to run tests. Tests the nesc1 driver in src/nesc1, except some stuff depends on the installation too (this is broken and should be fixed, do a recursive grep for ncc and nescc)

• src: the nesC compiler itself (builds nesc1)

nesc1 structure

nesc-* contain the nesC-specific parts of the compiler. The other files are related to handling C, but have a limited amount of nesC-related changes.

The main files are:

• toplev.c: contains main, parses options and invokes nesc_compile
• nesc-main.c: nesc_compile is the main entry point to the actual compiler
• nodetypes.def, AST*.c: nodetypes.def defines the types representing the abstract syntax tree (AST) that is the parsed representation of a nesC program. The AST*.c files are hand-written and automatically-generated files for creating and manipulating AST nodes. See nesc-internals/AST for more details.
• decls.h: declarations for types representing the various kinds of C entities (structures, fields, variables, functions, labels, environments, etc)
• nesc-decls.h: declarations for types representing the various kinds of nesC entities (interfaces, components)
• c-lex.c: the lexer
• c-parse.y: the parser
• machine.c, machine.h, machine/*: cross-compilation support
• cval.c, constants.c: constants and constant folding
• expr.c: C expression handling
• init.c: C initializer handling
• semantics.c: C declaration handling (structures, variables, functions)
• stmt.c: C statement handling
• types.c: C type handling
• unparse.c: print a C program corresponding to an AST - contains various hacks for nesC support

• nesc-component.c: code common to configurations and modules
• nesc-configuration.c: configurations
• nesc-module.c: modules
• nesc-interface.c: interfaces
• nesc-abstract.c: generic components
• nesc-atomic.c: atomic statements
• nesc-attributes.c: nesC @blah(...) attributes
• nesc-magic.c: magic functions (unique & co)
• nesc-network.c: nx_* types
• nesc-task.c: tasks

• nesc-cg.c: cgraph is used to represent component connection graphs and function call/use graphs...
• nesc-constants.c: perform a constant-folding pass
• nesc-deputy.c: support for Deputy's type-safety annotations
• nesc-msg.c: mig support
• nesc-cpp.c: interaction between nesC and the C preprocessor
• nesc-xml.c, nesc-dump.c, nesc-dfilter.c, ND*, nesc-dspec.def: dump information on nesC program in XML (the .def and automatically-generated ND* files reuse the AST infrastructure)
• nesc-doc.c: TinyOS 1.x documentation generator (obsolete, doesn't support generic components)
• nesc-ndoc.c: documentation string handling
• nesc-generate.c: nesC code generation (builds on unparse.c)
• nesc-inline.c: decide what to inline

Main Types

See nesc-internals/AST for a discussion of the abstract syntax tree representation. The abstract syntax tree closely matches the textual structure of the program, the types described next represent its logical structure, and are pointed to from AST nodes as appropriate (e.g. an AST node representing the use of a variable 'x' contains a field that points to the common data_declaration object that logically represents 'x' to the nesC compiler).

These types are all defined in decls.h, types.h, nesc-decls.h and nesc-cg.h.

C has three namespaces, one for symbols (variables, functions, typedefs), one for tags (structure, union and enum names) and one for labels. nesC adds a namespace for interfaces and components. This is reflected in the types below.

environment: a C scope, with definitions for symbols and tags.

data_declaration: a C symbol, with lots of information. AST nodes that declare or use a given symbol have a field (usually called ddecl) that points to the common data_declaration object for the symbol.

tag_declaration: a C tag. Like symbols, AST nodes and types that declare or use the tag have a field (usually called tdecl) that points to the common tag_declaration object.

field_declaration: a C field. As usual, appropriate AST nodes have a field (usually called fdecl) that points to the common field_declaration object. The field_declarations can also be found starting at their containg tag_declaration.

label_declaration: a C label. As usual, appropriate AST nodes have a field (usually called ldecl) that points to the common label_declaration object.

nesc_declaration: a nesC component or interface. As usual, appropriate AST nodes have a field (usually called cdecl) that points to the common nesc_declaration object.

type: an abstract object representing a C type. These objects are manipulated using the functions declared in types.h.

cgraph: a graph between nesC endpoints (endp). An endpoint is a triple of a component, interface and function. The endpoint also specifies any arguments for parameterized interfaces. The component and interface are both null for global C functions, the interface is null for functions internal to modules, and for commands and events that are not in interfaces. cgraph is used to represent component connection graphs and call/use graphs.