By early 1984, most of the components of the SERC Software Common Base were available or close to release. Most of the activities in 1984 were concerned with improving them in the light of comments. The compilers were certainly not as good both in compile time speed and quality of code generated as they might have been. Luckily, the compilers available on the main competitors (SUN and APOLLO) were equally bad as far as quality of code was concerned. However, any enhancements in this area would be a significant advantage.
The Window Manager and SPY editor were reasonable development vehicles. Various suggestions for change had come from the user population and these were worked upon in 1984.
Finally, there was a clear need for a set of basic facilities on top of the window manager to allow users to construct interactive programs without having to worry about how pop-up menus were handled and similar graphical tools. This set of facilities was given the cryptic UNIX name of WW.
ICL had decided in 1983 to produce an optimised FORTRAN compiler for PNX similar in performance to the one produced by ERCC for the POS operating system. It was to be a joint activity with Three Rivers.
ERCC gave a presentation to ICL and myself (representing SERC) near the end of 1983 of what they could provide. Based on their F77 work for POS and the 2900 series, they would produce a FORTRAN similar in functionality to POS FORTRAN but with the optimisations included in the 2900 compiler. This was not on general release at that time but early tests indicated it was showing good performance improvements over the unoptimised version. Optimisations included rearrangement of arithmetic code generation, DO loop expansion, improving the microcode for FORTRAN, etc. Optimisations not included were strength reduction on exponentiation, loop unrolling and a complete test for common sub expressions. The estimate given was a compile speed of 2000 lines per minute with a 25-40% improvement of execution speed over the existing FORTRAN compiler.
TARTAN Systems, a software house derived from CMU, also put in a proposal. This was a much more grandiose proposal incorporating a complete set of compilers with a high level of optimisation. Much of what they proposed is commercially confidential.
ICL asked a number of people to assess the rival proposals and give them guidance. The general view was that TARTAN had the better product in general but their knowledge of FORTRAN was not enormous and the safer path was to contract with ERCC to produce the compiler.
ICL accepted the recommendations and ERCC spent 1984 developing the new FORTRAN compiler and it was made available with the release of PNX4. Early assessment suggests that this provides a significant advance over the original compiler.
GKS was on field trial by March 1984 with PNX2.0. It proved to be a very stable implementation with a closeness to the ISO specification that was lacking in a number of alternative products. It was checked out against the GKS validation suite being put together at that time by Leicester, Darmstadt and GMD and uncovered as many errors in the validation suite as it did in the ICL/RAL implementation. The quality of this implementation in terms of specification was much dependent on the expertise at RAL, due to RAL being the document holder for the ISO standard and editors of the final ISO document.
Being a portable version, the early releases were slow and work was needed to tune the implementation to the PERQ. Some work was done late in 1984 but, even in 1985, work was still required.
A preliminary release of the NAG Library appeared with PNX1.5. A full implementation came with PNX2. 0 and provided; a much improved base on which to mount applications software.
The basic window manager was enhanced by a library of routines written at RAL. The aim was to provide an application interface to the user which shielded him from the differences between the various window managers on commercial systems.
It provides a set of basic graphic primitives to be used within a window management environment. For example, various fonts can be used for textual output, simple line drawing relative to a window, support for rasterop functions, subdivision of windows into boxes, texturing, event input, cursor specification and pop-up menus.
Interfaces are provided to FORTRAN, PASCAL and C. The aim is to have the library available on all systems in the Common Base and possibly Whitechapel MG-1.
