Documentation for the rest.

doc/scsh-manual/threads.tex

@@ -2,17 +2,17 @@
 
 \chapter{Concurrent system programming}
 
-The Scheme Shell provides you with support for concurrent programming.
-Its interface for concurrent programming consists of several parts:
+The Scheme Shell provides the user with support for concurrent programming.
+The interface consists of several parts:
 \begin{itemize}
 \item The thread system 
 \item Synchronization vehicles 
 \item Process state abstractions
 \end{itemize}
-Whereas the user deals with threads and synchronization explicitly, the
-process state abstractions are built into the rest of the system
-transparent for the user. Section \ref{sec:ps_interac} describes the
-interaction between process state and threads.
+Whereas the user deals with threads and synchronization explicitly,
+the process state abstractions are built into the rest of the system,
+almost transparent for the user. Section \ref{sec:ps_interac}
+describes the interaction between process state and threads.
 
 \section{Threads}
 
@@ -30,7 +30,12 @@ The thread structure is named \texttt{threads}, it has to be opened explicitly.
 Create and schedule a new thread that will execute \var{thunk}, a
 procedure with no arguments. Note that Scsh's \ex{spawn} does
 \textbf{not} return a reference to a thread object. The optional
-argument \var{name} is used when printing the thread.
+argument \var{name} is used when printing the thread. 
+
+The new thread will not inherit the values for the process state from
+its parent, see the procedure \texttt{fork-thread} in Section
+\ref{sec:ps_interac} for a way to create a thread with
+semantics similar to process forking.
 
 \defun {relinquish-timeslice} {} \undefined
 
@@ -139,16 +144,15 @@ Scsh provides an synchronous interface to the asynchronous signals
 delivered by the operation system\footnote{Olin's paper ``Automatic
   management of operation-system resources'' describes this system in
   detail.}.  The key element in this system is an object called
-\textit{sigevent}, which corresponds to a single occurence of a
+\textit{sigevent} which corresponds to the single occurrence of a
 signal. A sigevent has two fields: the Unix signal that occurred and a
-pointer to the next event that occurred. That is, events are kept in a
-linked list in increasing-time order. Scsh provides various procedures
-to access this list, they are all procided by the structure
-\texttt{sigevents}.
+pointer to the sigevent that happened or will happen. That is, events
+are kept in a linked list in increasing-time order. Scsh's structure
+\texttt{sigevents} provides various procedures to access this list:
 
 \defun {most-recent-sigevent} {} {sigevent}
 
-Returns the most recent sigevent, that is, the head of the sigevent
+Returns the most recent sigevent --- the head of the sigevent
 list.
 
 \defun {sigevent?} {object} {\boolean}
@@ -158,7 +162,7 @@ The predicate for sigevents.
 \defun {next-sigevent} {pre-event type} {event} 
 
 Returns the next sigevent of type \texttt{type} after sigevent
-\texttt{pre-event}. If no such event exists, the procdure blocks.
+\texttt{pre-event}. If no such event exists, the procedure blocks.
 
 \defun {next-sigevent-set} {pre-event set} {event}
 
@@ -190,24 +194,44 @@ variable \texttt{state} by USR1 and USR2:
 \section{Interaction between threads and process state}
 \label{sec:ps_interac}
 
-Global process state is a bad thing: it undermines modularity. In the
-case of concurrency however things get even worse.  The simplest
-example for this it the current working directory. If this would be
-global state, no thread can ever reliably dereference a relative link.
-
-Scsh addresses the problem of process state in a uniform way for
-almost all resources. For every global resource there is a
-procedure \ex{with-}\textit{resource}\ex{*} \var{thunk} which guarantees that
-during the execution of \var{thunk} the resource is 
-set to the desired value. There is only one exception: The uid under
-which the current process is running. The superuser may change to an
-arbitrary user without being prompted for a password, but the way back
-is blocked.
-
-
-
-
-
+In Unix, a number of resources are global to the process: signal
+handlers, working directory, umask, environment, user and group ids.
+Modular programming is difficult in the context of this global state
+and for concurrent programming things get even worse. Section
+\ref{sec:event-interf-interr} presents how scsh turns
+the global, asynchronous signals handlers into modular, synchronous
+sigevents. Concurrent programming also benefit from sigevents as every
+thread may chase down the sigevent chain separately.
 
+Scsh treats working directory, umask and environment as a thread-local
+resource. The initial value of the resources is determined by the way
+a thread is started: \texttt{spawn} assigns the initial values whereas
+\texttt{fork-thread} adopts the values of its parent. Here is a
+detailed description of the whole facility:
+
+\begin{itemize}
+\item The procedures to access and modify the resources remain as
+  described in the previous chapters (\texttt{cwd} and \texttt{chdir},
+  \texttt{umask} and \texttt{set-umask}, \texttt{getenv} and
+  \texttt{putenv}).
+\item Every thread receives its own copy of each resource.
+\item If \texttt{spawn} is used to start a new thread, the values of
+  the resources are the same as they where at the start of scsh.
+\item The procedure 
+
+\defun {fork-thread} {thunk} \undefined 
+
+from the structure \texttt{thread-fluids} starts a thread which
+inherits the values of all resources from its parent. This behaviour
+is similar to what happens at process forking.
+\item The actual process state is updated only when necessary, i.e. on
+  access or modification but not on context switch from one thread
+  to another.
+\end{itemize}
+
+For user and group identities arbitrary changing is not possible.
+Therefore they remain global process state: If a thread changes one of
+these values, all other threads see the new value. Consequently, scsh
+does not provide \texttt{with-uid} and friends.