Chair of Software Engineering

Eiffel: The Language -- Work In Progress
Bertrand Meyer

This revision: 5.00.02, 15 August 2005

To Latest news and changes

To Short table of contents and links to chapters

 Overview and restrictions

This private set of pages contains the current state of the book "Eiffel: The Language", third edition. All the material is copyright 1988-2002, Bertrand Meyer. (As you will see in the text, the copyright for the language reference and ELKS parts is meant to be released for free use once the book is in final form.) These pages are meant for consultation by colleagues and friends, especially members of the ECMA Eiffel standardization committee. To avoid lots of potential trouble, in particular with the publisher, please do not link to the actual PDF, only to the present page at http://se.ethz.ch/~meyer/ongoing/etl/, and do not make any publicly accessible copies.

Responsibility for the language now rests with ECMA-TC39-TG4, the technical committee that is standardizing the Eiffel language at ECMA. To participate in the committee you may join ECMA. The present pages serve as working material for the ECMA committee but have no official ECMA status.

 Status indications

The chapters are marked as being in one of the following three states:

"Not done": the chapter is a bare mechanical translation of the original troff text of the first edition. It usually has many inconsistencies, errors, missing elements and bizarre properties (even if some extracts appear reasonable). Do not trust anything in this chapter. In a few cases the chapter is just a placeholder.

"In progress": some work has been done on this chapter, but it is far from complete.

No particular mention: the chapter is in reasonably clean form, although in most cases it hasn't been seriously proofread.

Of particular interest is the "changes" appendix which lists the changes with previous versions. I try to keep it as up to date as possible.

 Latest news and changes

You can see the effect of these changes in the PDF chapters

Year 2005

15 August 2005: Version 5.02.00-0

Working now on the book: Standard Eiffel (ETL 3). In the process, making some small changes for future version of standard.

Chapters 2 to 9 have been processed. Rest hasn't been touched.

25 June 2005: Version 5.00.00

This is the ECMA standard version

12 June 2005: Version 5.00.00-4

23 May 2005: Version 5.00.00-2

Last fixes intregrating corrections by Alexander Kogtenkov and Kim Waldén.

21 May 2005: Version 5.00.00-1

This version has no change of substance but integrates corrections (from Kim Waldén, David Hollenberg and Alexander Kogtenkov).

6 May 2005: Version 5.00.00-0

This is the draft Eiffel standard as submitted to ECMA.

5 May 2005: Version 4.99.99-9

In this revision I have not changed the standard proper; I have, however, added an introductory section, lifted from "Eiffel in a nutshell" in ETL.

Change of mode of operation: I am no longer including diff files in the text. Instead there is a new (>45 MB) file ALLDIFFS.pdf which records all the successive diffs. Alternatively: http://tinyurl.com/dfoq2

This is the version to be submitted to ECMA Friday early morning. Please tell me of any typos that you find until then.

5 May 2005: Version 4.70.00-0

Improved aliasing rules; see discussion with Alexander Kogtenkov and Éric Bezault.

Added the notion of "non-production syntax rule" throughout. Added such a rule for assertion clauses.

Diffs added at the end.

Note that the FrameMaker diffs shows an entire chapter as different towards the beginning. This is obviouly a bug; just ignore that part. There actually only a few differences.

4 May 2005: Version 4.69.00-0

Integrates exceptions. See last diffs.

3 May 2005: Version 4.68.00-0

Improved presentation of the role of aliases (Feature_name vs Extended_feature_name). No new diffs; see message to the list.

3 May 2005: Version 4.67-02.0

3 May 2005: Version 4.67.00-1

Fixed (I hope) call syntax in Partition 23. No other changes, and hence no new diffs.

3 May 2005: Version 4.67.00-0

Equality/duplication partition ready (23). I produced no diffs since nothing else has changed. The last two diffs are still there.

2 May 2005: Version 4.66.01-0

Integrating Manu's remarks on version 4.65.00-0

1 May 2005: Version 4.66.00-0

See diffs (again only one, to 4.64, I removed 4.65<-4.64 diffs since no one commented on them).

30 April 2005: Version 4.65.00-0

See diffs (please check!). I removed all the old diffs so there's only one.

24 April 2005: Version 4.63.00

See diffs at the END

23 April 2005: Version 4.62.00-0

See diffs, included at the end of the document.

22 April 2005: Version 4.61.00-0

Processed corrections on constant chapter. PLEASE SEE FILE
DIFF4-57.pdf)

22 April 2005: Version 4.60.00-0

The second part, starting page 115, of the standard document

http://www.inf.ethz.ch/~meyer/ongoing/etl/(see here)

is the set of diffs with 4.58. PLEASE REVIEW IT. Please ignore the differences in the front matter (already present in 4.59, but there was no diff), already reviewed, and examine the rest of the document.

22 April 2005: Version 4.60.00-0

Please see DIFF file. Introduced notion of generic-creation-readiness and its consequences.

17 April 2005: Version 4.59.00-0

To review this version, please consider only the front material. I have completed it except for the Eiffel introduction in 8.1.

10 April 2005: Version 4.58.00-0

Continuing to implement changes discussed in phone meetings. See diffs (58 <- 57) at end of text. Previous diffs (57 <- 55) also included, just before thæ latest one.

10 April 2005: Version 4.58.00-0

Continuing to implement changes discussed in the meetings. See new diff file at the end (Previous diff, 57 <- 55, also included, just before that one).

9 April 2005: Version 4.57.00-0

9 April 2005: Version 4.57.00-0

Implementing changes and suggestions. See diff file: (see here)

3 April 2005: Version 4.56.00-1

Diffs produced (see end of standard text))

Processed Eric's round 3 remarks on front matter and chapters 2 to 4.

(Round 1 -> Eric's first comments on full text, 2-column text Round 2 -> His new comments on some chapters, 2-column text Round 3 -> First pass on final-format text, after first review)

Processed Round 1 remarks and consequences of last phone discussion on chapter 5 (features)

30 March 2005: Version 4.55.00-1

Integrating comments on chapter 2, 3, 4 after discussion with Eric and Manu

28 March 2005: Version 4.54.00-1

Diff file regenerated (still diffs with 4.53).

Full pass through the book to underline defined terms in rules and definitions.

Used the opportunity to improve a bit the rules about genericity and several others (e.g. "Converting to and from a type" in the conversion chapter).

Changed "conversion function" to "conversion query".

27 March 2005: Version 4.54.00-0

Text to be checked: (see here). Contains diffs at end.

Revised creation and assignment chapters. Processed all of Eric's comments except those relating to assertions (in creation).

Removed assignment attempt stuff.

Simplified and improved Assigner_call description.

First attempt at reattachment semantics; this is incomplete and needs checking.

22 March 2005: Version 4.53.02-0

Almost no changes of substance, but I continued to fine-tune the format for the standard.

I brought the "Explanatory text" inserts up to and including the Routine chapter.

Next release will be next week-end.

20 March 2005: Version 4.53.01-2

Continued update to document in standard form (see here)

Started removing assignment attempt (syntax and validity only so far, for standard)

Note that individual chapters are not up to date with my last changes due to a network problem at ETH (I hope the system administrators will fix it on Monday). The standard text (see here) is, as well as the full book text (see here)

20 March 2005: Version 4.53.00-1

No change to substance, but this version now includes a form closer to what ECMA requests.

19 March 2005: Version 4.53.00-0

NEW DIFF FILE DIFF4-53.pdf

Second pass on expression chapter (see here).

In syntax variants, replaced Formal by Read_only (otherwise Current was not an expression!).

Integrated Eric's comments:

I disagreed with just one: for VWOE there is in my view no need for a condition on the number of arguments. This is taken care of by the Equivalent Dot Form (see comment in text).
Added Once_string to VWBR but not Precursor because of the "Hmmm".
Changed order of clauses in definition of expression type, so that it follows order of syntax.

18 March 2005: Version 4.52.05-0

More corrections on (see here) based on Eric's comments

18 March 2005: Version 4.52.04-0

More fixes to call chapter (see here)

17 March 2005: Version 4.52.02-0

NEW DIFF FILE DIFF4-52-2.pdf)

Changed colors in standard text -- please comment if you have better ideas.

Introduced new convention in response to requests: terms used in a definition or rule are doubly underlined if they are themselves formally defined elsewhere. The double underline is a bit heavy but I didn't find anything better so far. Single underline makes one want to click; italics conflicts with other uses of (colored) italics; I tried using a different font but it's ugly; and FrameMaker doesn't have dotted underline, which would be my preferred convention.

Continuing revisions of (see here):

Simplified target syntax (Eric's suggestion)
Changed unfolded form of Non_object_call.
Rewrote semantics of routine calls to factor out assertions and take care of conversion before conformance (the latter was not done before, causing wrong semantics).
Adapt some of the definitions in a simple way to remove cyclic dependencies (there seems to be cycles, but see the note on page 552).

In this version I did not change the order of topics (current object, target object etc.); I'll do it later tonight if I have the time, otherwise I will just give up for the moment.

In type chapter (see here):

Replaced "unfolded form of an anchored type" by "anchor-free form of a type", applicable to any type. This is useful in particular for the unfolded form of a Non_object_call.

16 March 2005: Version 4.52.01-0

NEW DIFF FILE

Corrections to call chapter (see here) after this morning's discussion

Changed Object_test to have an Unqualified_call rather than a Call_chain.
Added Object_test to variants of Special_target.
Merged the semantics of qualified and unqualified calls.
Added (for call expressions) definition of "Result of a call" and "value of a call expression".
Defined more precisely the evaluation of assertions in a call.
Explicitly mentioned that compilers may skip target validity checking (but then the result is invalid). This may have to go to the Void Eradication chapter.

15 March 2005: Version 4.52.00-0

**NOTE DIFF FILE: (see here)**

The goal of this version is to conclude the call chapter (see here) (as the aim for 4.51 was to conclude (see here)).

On the call chapter:

Took into account Eric's remarks following this morning's discussion.
Improved the definition of "feature of a call" to account for multidot calls in a non-ambiguous way.
Significantly simplified Argument rule, so that it involves only the static type. Part about descendant now moved to type checking chapter (Descendant Argument rule, see below).
Added references to exceptions and assertions. Didn't update once stuff yet.

In the type checking chapter (see here):

Added Descendant Argument rule, with detailed explanations (the explanations don't affect the Standard, of course).

In the exceptions chapter (see here):

Moved the chapter to after calls. Removed the separate chapter on "exception facilities" and brought the material (out of date) here.

15 March 2005: Version 4.51.00-0

New diff file: (see here)

Fixing expression chapter (see here) after this morning's discussion:

Boolean expressions: fixed syntax (constants, Basic_expression).
Finalized notion of type of an expression, added missing cases.
Address is now `$' Variable.
Numerous small changes and corrections.

In call chapter (see here):

Introduced notion of Special_target to correct missing possibility of local, formal etc. as target of a call (see email correspondence with Eric).
Improved the presentation of Non_object_call by explicitly introducing an unfolded from.

In feature chapter (see here):

Updated Alias Validity rule to ensure that the semistrict operators can only appear in BOOLEAN.
Formally defined the notion of synonym.

In redeclaration chapter (see here):

Took care of precursor, defining unfolded form.

Worked on the front matter to prepare for appearance as a standard. The content is tentative; I will check with ECMA. Please check credits for any mistake.

12 March 2005: Version 4.50.01-0

Reference (see here) has been regenerated (but didn't check). No new diffs.

Introduced explicit conversion through feature of TYPE: ELKS chapter (see here),

In the expression chapter (see here):

Reorganized chapter, changing order of sections

Moved equality discussion to chapter on copy/equality chapter (see here).

Removed VWEQ constraint (see issue OPV).

Changed format of operator precedence spec so that it now appears in standard.

Started to take into account Eric's comments on the chapter. Need to handle Precursor (not finished yet).

Rewrote section on explicit conversion, taking advantage of feature in TYPE (see above).

Working on the attached mechanism. Changed:

- Entities (20.6, VEEN)

- New section 28.8 on Object Test

- New chapter in progress: 25 on eradicating void calls void.pdf.

11 March 2005: Version 4.49.03.0

In the process of documenting the attached stuff. Intermediate version, but I would appreciate if someone could look at the definitions on pages 654-655. Note that the reference has not been properly regenerated.

11 March 2005: Version 4.49.03.0

In the process of documenting the attached stuff. Intermediate version, but I would appreciate if someone could look at the definitions on pages 654-655. Note that the reference has not been properly regenerated.

11 March 2005: Version 4.49.03.0

In the process of documenting the attached stuff. Intermediate version, but I would appreciate if someone could look at the definitions on pages 654-655. Note that the reference has not been properly regenerated.

11 March 2005: Version 4.49.03.0

In the process of documenting the attached stuff. Intermediate version, but I would appreciate if someone could look at the definitions on pages 654-655. Note that the reference has not been properly regenerated.

6 March 2005: Version 4.49.02.0

Continuing to update genericity chapter. Introduced "frozen" formal generic parameters (they were partly there but not in the syntax). Started to remove the special case of unconstrained genericity (as suggested by Kim in a Zurich meeting) and hence VTUG. Will finish this tomorrow. Moved reference appendices to the very end, with their own numbering. There's no need to waste time checking this version, it's intermediate. No diffs.

6 March 2005: Version 4.49.01-2

5 March 2005: Version 4.49.01-1

Corrected a few mistakes pointed out during tonight's phone meeting. No new diffs generated.

5 March 2005: Version 4.49.01-0

Revised type and genericity chapters to take into account last night's discussion.

Introduced Class_or_tuple_type (see email discussion). Some rather delicate changes, in particular to definition of base type of a formal generic.

3 March 2005: Version 4.49.00-0

Integrating all of Eric's comments and the discussion of last night. I am only listing here the comments NOT taken into account yet.

Conversion type in definition of simple client: not integrated.

Did not yet update terminology replacing "short" etc.

Only a few updates on assertion chapter.

Removed the discussion of "static type" at the end of 10.26 since there is a better definition in the chapter on expressions (28.13; Ialso improved it).

Diffs with previous version: DIFF4-49-00(see here)

Earlier diffs: DIFF4-48-05(see here)

2 March 2005: Version 4.48.05-0

Worked on ELKS chapter (see here).

Updated TYPE; replaced is_same_type by redefinition of is_equal; added `adapted alias []'.

Removed all `require other /= Void' preconditions and most `ensure Result /= Void' postconditions; added a few ? to arguments.

The good news is that I found only one need for a ? (in STD_FILES). But all arguments need to be checked to see if more could legitimately be void and hence require a ?.

To all "-" functions, added a postcondition Result + other = Current This may not work for REAL types and if so will be replaced by a comment.

General cleanup, put assertion tags in roman, replaced "note" by "comment" when used as note tag.

Replaced all "empty" features by "is_empty"

Added INTERVAL class, with PART_COMPARABLE generic parameter. Many classes now have an alias ".." function that returns an interval; as a consequence ".." is a standard operator. (This is largely the consequence of a suggestion by Marcel Satchell. It seems right but I can roll back if the change turns out to be a bad idea.) An ARRAY now has an associated interval. One of the advantages of having INTERVAL and the associated ".." operator is the ability to write assertions such as (|1..count|).for_all (...).

No change to language reference.

There will probably be another version later today (meaning tomorrow Zurich time).

2 March 2005: Version 4.48.04-0

Small changes only, because of tool problems; modified lexical chapter (see here) to integrate comments by Marcel Satchell.

2 March 2005: Version 4.48.03-1

1 March 2005: Version 4.48.02-4

1 March 2005: Version 4.48.02-2

Fixed PDF generation problem (I hope).

1 March 2005: Version 4.48.02-1

Checking generation, version published on the Web doesn't seem right.

28 February 2005: Version 4.48.02-0

Starting the final run towards submission of the standard; integrating the (many) decisions made at the last meeting and the preceding ones. For the moment:

Target expressions (new section 24.6); used this opportunity to do some cleaning in the Call chapter (see here).

Unfolfed form of a multiple declaration (5.18).

Rules on feature declarations (5.20), in particular aliases. There is a new "alias validity" rule that also applies to renaming.

Discussion of renaming

Started to explain assigner procedures

Removed most references to the earlier version of the "attached" mechanism.

In addition, there is now a "diff" for the Language Reference, comparing it to the previous version. This comes just after the first version of the reference (currently appendix K, see here).

25 February 2005: Version 4.48.01-0

Working on constant chapter

20 February 2005: Version 4.48.00-2

Incremental improvement, working on constant chapter (see here)

20 February 2005: Version 4.48.00-1

Changed title of book.

19 February 2005: Version 4.48.00-0

Finished lexical chapter (see here). Proofreading will be appreciated!

19 February 2005: Version 4.47.05-0

Contrary to what was written earlier, I have gone on to the lexical chapter (see here). Should be finished soon.

18 February 2005: Version 4.47.04-1

18 February 2005: Version 4.47.04-0

Basically finished the conversion chapter with new rules regarding preconditions; still need to proofread and clean up the chapter.

17 February 2005: Version 4.47.03-1

Finishing conversion chapter to integrate all recent decisions, in particular the rule on preconditions allowing us to convert freely between natural, integer etc. Will get back to tuples immediately afterwards.

16 February 2005: Version 4.47.02-2

Took care of manifest types in expressions. It is now possible to have several of them, e.g. {V} {u} {T} exp. This doesn't cause any ambiguity.

17 February 2005: Version 4.47.02-1

Added a syntax description convention: X* is a plain repetition of X, without delimiter; same for +. Simplified syntax in a few places thanks to that convention.

15 February 2005: Version 4.47.02-0

Working on the tuple chapter, fixed obsolete elements integrated recent decisions.

-- Change log lost for January and early February versions --

Year 2004

5 December 2004: Version 4.46.00-0

Finished the "entity" chapter and the specification of lazy semantics for attached and expanded types.

Started to remove references to DOUBLE (partially only, the only advantage of Frame 7.1 was the ability to search through a full book).

4 December 2004: Version 4.45.02-0

While continuing on semantics of expanded, entities etc. (significant changes to chapters on types and entities), updated syntax for non-conforming inheritance. Note that the result is slightly different from what Eric has proposed; the reason is that it's much simpler that way. Obviously this will need to be discussed.

4 December 2004: Version 4.45.01-0

Removed "Type_descriptor as one of the variants of Agent_actual to avoid a syntactic ambiguity. This corresponds to a decision of the committee, see 10th meeting, 20-22 June 2004, 14.4 see page of all_minutes.pdf.

Removed the notion of "Remote_constant", now covered by Non_object_call.

28 November 2004: Version 4.45.00-4

Continuing on the new semantics of expanded.

28 November 2004: Version 4.45.00-3

28 November 2004: Version 4.45.00-1

Currently adapting to the new expanded semantics and removing the notion of expanded type (other than for types based on expanded classes). This results in significant simplifications, in particular the removal of VTET (VTEC in ETL 1/2), now replaced by a new clause 2 in VCCH.

20 November 2004: Version 4.45.00-1

Not much change, but regenerating text now that I have reversed to FrameMaker 5.5 (great relief to have software that works)

31 October 2004: Version 4.45.00-0

Finished the call chapter. Also started to update text to take into account lazy semantics for attached and expanded (see "Entity Semantics").

17 October 2004: Version 4.44.02-0

Due to a mistake on my part the log for all versions since April has been lost. Too bad since there has been extensive activity recently. I am in the middle of in-depth modifications; took care of assignment procedures, operator notation, bracket notation, attached types (not finishes), object test, attached clauses in loops, call validity and semantics etc.

17 October 2004: Version 4.44.02-0

5 April 2004: Version 4.42.00-2

I now have a more convenient system for automatically notifying committee members of updates, both by email and on the Web page. I developed it yesterday; of course some glitches may remain. I hope to find enough time in the next few weeks to make good progress and start taking advantage of all the committee's work in recent months, especially the last meeting. Thanks for very productive discussions.

4 April 2004: Version 4.42.00-01

Setting up files for the forthcoming updates

Year 2002

21 Sept 2002: Version 4.42-00-00

Following an informal meeting on 16 September and a suggestion by Éric Bezault, added rule that if there is an inheritance part at least one of the Parent clauses must specify conforming (non-expanded) inheritance. This ensures the property that every type conforms to ANY, which otherwise would disappear as a result of the introduction of non-conforming inheritance. That property is now stated as a rule, the Universal Conformance rule, which is not a new validity constraint but a theorem ensured by the other conformance rules. The affected sections are 6.6 and 6.9, both in the inheritance chapter.

I might rename "non-conforming inheritance" as "expanded inheritance" since this is what people tend to use.

I started some time ago updating the text to take into account the changes decided at the Santa Barbara meeting in August (contracts on attributes etc.). It's actually rather delicate since it affects many parts of the text and puts into question the very structure of the exposition, so it's going to take considerablee time.

11 June 2002: Version 4.40-00-11

(Updated here 10 August 2002, sorry for the delay). Although not reflected on this page, there has been much activity recently with the start of the ECMA standards group for Eiffel (ECMA TC39-TG4). Contact ECMA about participation. The version updated here corresponds to the initial text discussed in the June TG4 meeting.

Year 2001

31 December 2001: Version 4.40-00-00

Two significant language changes (hence the top-level version increment):

Added considerable generality to the once mechanism, allowing once keys. This is in the routine chapter and, mostly, in a section of the call semantics chapter (that's the only part of the chapter worked on so far). This has also meant changing the syntax of Debug since there is now no more notion of Debug_key and Debug_key_list, simply Key_list used for both Once and Debug.

Removed the assignment attribute mechanism altogether. This turned out to be too messy for its sake.

Other changes: chapter on older changes released; moved the long tutorial to the end of the book. Corrected and updated the recently done chapter on copying and equality and chapter on control structures to integrate comments by Marcel Satchell.

Finally, I updated the change appendix, the ELKS appendix and the long tutorial to reflect all the recent changes.

29 December 2001: Version 4.39-00-00:

New pass on chapter on control structures (see also the changes made one year ago). Worked on multi-branch to add the possibility of discriminating on strings and on types. Forward references to the planned change in exception mechanism, to be added soon. Cleanup and style improvements on the whole chapter. There will be another update today or tomorrow as I have since then updated the tutorial and "Eiffel in a Nutshell".

Lengthened the version number convention to give myself more opportunities of self-pats in the back for minute improvements. The convention (only slightly adjusted, and remaining compatible with previous version number) is:

Version number is 4.AA-BB-CC

When 4 turns to 5 the book is ready for publication.

AA (currently 79) is incremented for any major language change or addition.

BB is incremented for any small language change or addition.

CC is incremented for changes and additions to the text without significant language change.

8 December 2001: Version 4.38-1:

Finished chapter on copying and equality. Changes include making equality symmetric and correspondingly requiring copy to take an argument of type identical to its target, and introducing infix "}={" as a synonym to equal. As usual, many improvements (I hope they are improvements) to the explanations and the examples. I didn't touch the semantics of deep equality, although Michael Schweitzer tried to convince me a few years ago that it was brain-damaged; I will ask him if he remembers what the problem was.

18 December 2001: Version 4.37-3:

Files copied to ETH Zürich, everything is ready so that I can start working on the text again. Checking some issues with the creation mechanism (on expanded) based on comments by Emmanuel Stapf.

To list of changes for 2000.

To list of changes for 1999.

Year 2000

26 December 2000: Version 4.37-1:

I am working on the Lace chapter. This will take a while, so there will be a number of 4.37-x. Until further notice don't bother to look at it, it's still in a mess with text processing problems.

25 December 2000: Version 4.36-3:

Revision and improvement of the class naming mechanism, following comments by Éric Bezault, Roger Browne, Alexander Kogtenkov and James McKim.

19 December 2000: Version 4.36-0:

Addition of a mechanism to remove class names conflicts through qualification by cluster names. The chapter principally affected is the architecture chapter. Here is the message to the Talkitover group explaining this change:

This version adds a simple mechanism for removing name clashes between classes. ISE Eiffel (I don't know about others) had previously used, since 4.0 or so, a Lace renaming mechanism, but this has proved to be unsatisfactory. The policy that has been adopted now results from careful evaluation of those in other languages, in particular namespaces as present in recent versions of C++ and Smalltalk, as well as the .NET model and C# (requiring, in the latter case, the use of a keyword `using' similar to Ada's `with'). We found that this was overkill and adopted a much simpler solution. Within a cluster, no two classes may have the same name. A conflict arises if and only if you try to use a class name C such that no class of that name exists in the current cluster (if one does, it will be the one selected), but two or more classes are called C in other clusters. Then, to refer to that class, you must qualify it by its cluster name: cluster1::C or cluster2::C. Even when not required through this rule, such a fully qualified name is now a valid form of Class_name, valid anywhere we need a class name (e.g. in a type name, or in a selective exports clause), with the exception of a class definition itself,

class C

where C must be unqualified.

The cluster name comes from the outside, typically an Ace file or equivalent. It is thus local to every system, although a library may come with a mini-Ace, to be included as is, and specifying a cluster name, so that systems using the library don't have to define their own cluster names from the library.

We haven't implemented this mechanism but feel that it's the right way to go -- simple, not overkill, and safe. I am not too keen on the :: symbol, but don't have a better suggestion now. (I'd originally used a single :, but a declaration of the form x: cluster_name:CLASS_NAME seems a bit strange.)

The changes are mostly in the last section of the Architecture chapter (currently chapter 3).

19 December 2000: Version 4.35-3:

Typographical improvements and more corrections on agent chapter. following comments by Marcel Satchell and Sam O'Connor (on the EiffelVision example). (There was a 4.75-2 earlier today.)

14 December 2000: Version 4.35-1:

A few corrections on agent chapter following comments by Marcel Satchell.

13 December 2000: Version 4.35-0:

New version of the Agent mechanism and the corresponding agent chapter.

I have integrated a new syntax and various simplifications for agents. Other than errors (which probably remain, since I haven't reviewed the text extensively) I hope this is a final version of the mechanism. It uses a keyword syntax and has a simple, general notation for inline agents.

(Note: I find no record here of Version 4.34-0, so I must have forgotten to update the Web page back in September. It was the first step towards this update of the agent mechanism.)

25 August 2000: Version 4.33-1:

Fixed a few typos in the just revised section on Verbatim_string in the expression chapter.

Updated the appendix on reserved words and special symbols to reflect recent changes, in particular by adding the keyword agent.

25 August 2000: Version 4.33-0:

Agent chapter: the agent mechanism has been resyntaxed and slightly revaliditied and resemantiqued. Full chapter up to date. Keyword-based syntax (no more tilde ~ and vertical bar |); complete unification between feature-based agents and inline agents; full power of anonymous routines. The possibilities are quite interesting. Before starting PDF regeneration I forgot to update the keyword and operator lists at the end of the book (~ and | are out, `agent' is in) but have done it on my working copy, so the PDF version will be OK the next time around.

Also: in the discussion of Verbatim_string in the expression chapter, refined the notion further (see earlier change) to allow for left-justified verbatim strings. As before the expression chapter is still not ready, but some parts are, including the relevant one for this change: the section on manifest strings, currently 26.18.

13 August 2000: Version 4.32-0:

Agent chapter: New syntax proposal. Chapter is not complete (stopped on page 677, mark should be clear) but there is enough for discussion. Unifies routine agents with inline agents.

21 July 2000: Version 4.31-0:

Expression chapter: Refined the notion of Verbatim_string to allow for a way of escaping the end-of-string marker. This takes care of a very extreme and rare case, and for that case adds 1.5 pages of text, two syntax constructs, and a validity constraint, so it's not attractive, but someone finally convinced me that any marker has to have an escape convention.

Note that the expression chapter is still not ready, but some parts are, including the relevant one for this change: the section on manifest strings, currently 26.18.

4 May 2000: Version 4.27-0:

Chapter on Repeated inheritance now ready. I have incremented the third-level version number since this introduces a major language change, discussed in the past couple of months: simplifying the semantics of repeated inheritance by removing the Select clause and relying instead on non-conforming inheritance. Although I didn't expect to do any more chapter-moving at this stage, I have shifted the repeated inheritance chapter to after the chapters on type and conformance because the rules rely extensively on conformance. It's actually better that way since types and generics are more basic concepts than repeated inheritance.

As a consequence I have also had to make fairly extensive changes to Feature adaptation chapter released two weeks ago. To make them easier to spot I am also releasing a change bar version of the Feature Adaptation chapter (which will be removed in a couple of weeks). The most significant changes start on page 328.

A minor addition has also been brought to the conformance/convertibility chapter (definition of "conformance path"). Several other chapters have been touched as well.

As usual, the changes appendix has been brought up to date.

4 May 2000: Version 4.26-0:

Chapter on Feature adaptation now ready. I have incremented the second-level version number since this is the first time ETL officially integrates the Precursor mechanism, even though it has of course been available in practice for quite some time. Other than Precursor, the only significant language change is the removal of the prohibition on redefining an External routine into an Internal one, or conversely. This used to have a justification with early conventions on external routines, but became no more than an implementation restriction with no conceptual justification.

21 February 2000: Version 4.25-1:

Revised chapter on interfaces to external languages to support new, clearer, more regular syntax (see 4.63-0). Considerably simplified syntax for C, C++ and DLL sublanguages. Better conventions; there is hardly any need for a % any more. No more brackets in sublanguage specifications. This was essentially the work of Emmanuel Stapf. (The trend was started Pierre Métras some time ago.)

21 February 2000: Version 4.25-0:

Revised chapter on interfaces to external languages to support new, clearer syntax. See 4.63-1.

17 February 2000: Version 4.24-0:

non-conforming inheritance

Perhaps incautiously, I have gone ahead and introduced the notion of non-polymorphic inheritance in all the previously written chapters. (The main affected ones are features and conformance/convertibility.

The bad news is that I feel pretty happy with myself; the impact on the presentation has been remarkably simple. None of the conformance or convertibility rules need to be changed! All I have had to do is include a new optional component in the syntax, and a number of explanations and extra examples. This is bad news because when I feel that happy it's usually a sign that I should have waited the proverbial night and that there is probably something rotten in the kingdom, which one of the distinguished members of this group will undoubtedly discover right away. Well, I am foolish enough to publish the material as is tonight. We'll see (after the usual 90 minutes or so, only about 75 when this message reaches you).

Note that the simplicity of the change comes in part from the use of the keyword expanded for what Jim Nelson called flat inheritance. Although the purpose of a language design is not primarily to make the scribe's task easier, I -- being the self-appointed scribe -- am finding the appeal of self-convenience hard to resist in this case.

Before I move to anything else I have to revisit the conformance rules on tuples, which I don't think are right although I don't yet know the proper fix.

February 2000: Version 4.23-0:

(A minor release, 4.60-5, a few days ago, was not documented.) This new version on or before 23:41 tonight PDT, i.e. 08:41 Paris meridian time) includes two language changes:

Following a comment by Marcel Satchell I have updated the validity constraint for convertibility (CNCP, section 15.13, subsection "Conversion rules", currently page 449) to account for a case that I had missed regarding generic types. This was a serious mistake and I am grateful he found it.



In the agent chapter I have added a notion of inline procedure agent parenthesized by `do'...`end'. Previously inline agents described functions. This change arose of discussions with participants in the "Advanced Software Development Using Eiffel" taught last week, particularly Michael Gacsaly. I found I couldn't really justify the lack of consistency in the handling of procedures/functions (instructions/expressions). The addition doesn't complicate things too much. I even managed to keep the chapter's page count unchanged at the price of some cheating. See sections 25.17 and 25.18 in the agent chapter.

Other changes: I fixed a number of typos (thanks to Marcel, Ian Joyner, Pierre Metras, Joachim Durchholz and others) and removed references to the "Balancing rule" (some may remain.

1 February 2000: Version 4.22-4:

A fairly major new release, integrating the recently discussed conversion mechanism. (There have already been 4.60-0 to 4.60-3 in the past couple of weeks.) Several chapters updated, in particular features (see section on target conversion) and conformance, the latter renamed "conformance and convertibility" and now almost 50% longer than before, which is of course subject to criticism. Chapter on reattachment and equality is complete. Also, corrections on a number of other chapters, thanks to comments by Ian Joyner.

January 9, 2000: Version 4.21-3:

Chapter on objects, values and entities now ready. I have also used the opportunity to simplify and improve the type chapter and fill in some forgotten elements in the changes appendix. No new language changes.

January 6, 2000: Version 4.21-2:

attribute chapter now ready.

There is no language change in this release, except that I have included (in the semantics rule) the change indicated previously (see the entry on 4.21-0 just below) that guarantees uniqueness of Unique values even across classes. This change raises potential implementation problems so it might be rescinded later on.

Otherwise this is just a cleanup of the attribute chapter.

January 3, 2000: Version 4.21-1:

Corrections on type chapter (typos and inconsistencies) thanks again to Marcel Satchell.

January 2, 2000: Version 4.21-0:

Control Structures chapter now ready. Only one language change: the last clause of COMB, the constraint on Multi_branch instructions (clause 6 in ETL 1/2) has been removed. It required all Unique constants of a given Multi_branch to come from the same class. It is now felt that this is unfairly passing an implementation issue on to programmers. If there is too much complaining from implementers, however, the clause will be reinstated.

This version also includes numerous corrections on earlier chapters, courtesy Marcel Satchell's careful reading.

January 2, 2000: Version renumbering for consistency. The last few versions should have been called 4.20-x. They now are. The record (below and in the Talkitover archive) has been corrected. Sorry for any confusion.

Year 1999

December 30, 1999: Version 4.20-3.

This version corrects a number of errors in the agent chapter as a result of comments by Marcel Satchell (many thanks). Some of these had consequences on the ELKS chapter.

I have sneakily added two words at the bottom of page 630 which permit ~a where `a' is an attribute, the agent being of type FUNCTION (there was a reference to a class ATTRIBUTE but I have removed it -- too complicated). This possibility may be removed later, however, if it is found that the implementation hardship is not worth the extra benefit. (This would contradict the Uniform Access principle, but only for one specific language construct.)

December 29, 1999: Version 4.20-2

(There was an intermediate, undocumented 4.20-1 a few days ago.) This is a fairly major release. Here is some of what's in it:

Changes appendix: Updated to reflect essential changes, including those below.

Agent chapter: many small changes and improvements. Added creation agent example (application of inline agents). Changed feature name: value --> item

Agent chapter: introduced class PREDICATE (see also next entry) with convention that BOOLEAN-returning FUNCTION agents are instances of this class. Very important for work on trusted components (neither space nor time to explain here, but believe me).

Agent chapter and ELKS appendix: completed specification of ROUTINE, PROCEDURE, FUNCTION and PREDICATE. I hope it's close to final. Very close to current ISE version (this is essential: no mechanism should be definitively accepted until there is a working implementation); only major difference is that I have changed `argument-' feature names to `operand-' (have to check this with experts next week). I haven't done anything on class TYPE yet.

ELKS appendix and elsewhere: removed class GENERAL. ANY is the sole common ancestor of everything.

ELKS appendix: Added feature `twin' in ANY. Same effect as `clone' (and, like `clone', frozen), but works on target.

ELKS appendix: Changed terminology for `equal', `clone', `copy' and `twin' variants: `identical' instead of `standard_is_equal', `identical_clone' instead of `standard_clone' etc. In actual libraries the old ones should of course be kept as obsolete.

ELKS appendix and elsewhere: INTEGER, REAL types.

ELKS appendix: Removed BIT types. See boolean operations on integers.

Expression chapter: Defined the notion of type of an expression (wasn't there before!). Please read that section (currently 26.21 -- number will almost certainly change).

Type chapter and conformance chapter: Anchoring to formal generics and tuples now officially supported.

Expression chapter: String semantics: notion of `once' string, section (currently) 26.18. See Talkitover message explaining this in detail.

Conformance chapter: chapter complete. See in particular the new notion of Class_type_reference, conformance for multiply constrained generics etc. (Note: multiple constraints have not been implemented yet.)

Genericity chapter: Chapter complete. Many new definitions. Current type, etc.

December 17, 1999: Version 4.20-0.

Intermediate version -- conformance chapter is in progress, a few corrections elsewhere. The significant revision will come in the next few days. Working on several things (integer classes, conformance, anchored declarations).

December 12, 1999: Version 4.17-1.

Same as 4.47-0 but everything put back together on Unix; links should work normally.

December 4, 1999: Version 4.17-0.

New version of agent chapter integrating the new inline agent mechanism, and new examples (once per object, controlled execution). Also, class TYPE now has a generic parameter (this seems surprising at first but I think it's right -- see the text). Important caveat: the text has been prepared on Windows rather than Unix and will show a few glitches. In particular links to other chapters probably won't work. Page numbers are wrong (book has not been regenerated). This will only be fixed around Dec. 15.

November 5, 1999: Version 4.16-4. Changed creation expression syntax again in creation chapter. This is basically to undo the previous change (September 2) after some people convinced me it was a bad idea. So the previous syntax with the create keyword is back. I may have forgotten to update some of the examples; if you notice any please tell me. Also did some minor cleanup on the rest of the chapter.

November 3, 1999: Version 4.16-3.

Cleaned up system architecture chapter. The new "root procedure rule" (CSRP, page 165) was included some time ago together with the revised Root Class rule (CSRC, page 164), but I added the clause (obvious, but missed until now) that the root procedure must be a creation procedure of the root class. Updated changes appendix accordingly. Otherwise, no major change.

November 3, 1999: Version 4.16-2.

Cleaned up classes chapter. Nothing earth-shattering. Next is architecture chapter, so that the beginning of the book (except tutorials) will be all in a clean state.

November 1, 1999: Version 4.16-1.

Revised genericity chapter in line of conventions adopted for type chapter (see 4.46-1). Made small adaptations to type chapter to mention that for Formal_generic_name expanded/reference status is known only for a specific generic derivation.

October 31, 1999: Version 4.16-0.

Finished type chapter. Relaxed VTAT rule on anchored types so that it is permitted to use an anchored entity as anchor; this requires a rule on the absence of cycles. (See Michael Schweitzer's note on the topic in the VIP discussions; note that I have not found necessary to follow his suggestion for a similar no-cycle rule on generic constraints.) A small point of semantics has been left unaddressed (reference/expanded status is not known if base type is Formal_generic_name). I will take care of this at the same time I clean up the genericity chapter to make sure that type semantics is uniformly in the same style: specify reference or expanded status; specify base type.

Updated changes appendix to reflect language update.

October 23, 1999: Version 4.15-1.

Worked on type chapter. Systematized the presentation, especially for semantics. Defined the semantics of the `reference TYPE_NAME' types (by symmetry with `expanded TYPE_NAME') as proposed by Franck Arnaud. Stopped after page 390; anything after that is in an unsuitable state. For the "finished" part up to page 390, the usual caveats apply. In terms of work of substance rather than form, the serious piece of work that remains to be done on this chapter is to allow anchoring a type, as in `like t', on an expanded `t'.

October 22, 1999: Version 4.15-0.

Simplified syntax for multi-line strings, now called "verbatim strings": "[ and ]" rather than "[% and %]". Updated corresponding section in expression chapter (but rest of chapter still unfit for human consumption). Updated examples in externals chapter accordingly. Started work on type chapter. (I know, this is not the one I was supposed to handle now, but it shouldn't take long and it felt natural.) Usual caveats apply.

Updated changes appendix to reflect language update.

October 17, 1999: Version 4.14-5.

Improved syntax of assignment procedures in feature chapter, following a suggestion of Roger Browne. Finished Syntax, validity and semantics chapter. Usual caveats apply.

October 10, 1999: Version 4.14-4.

Implemented new version numbering system. Updated DLL syntax to allow for names as well as indexes. Touched a few other chapters (structure, basic conventions, regarding letter case).

October 9, 1999

For the first time, brought the appendix on reserved words and special symbols to a clean state, although I don't claim completeness and haven't introduced the cross-references yet.

October 9, 1999

Made a change in the C interface sublanguage following another of Pierre Métras's suggestions: using a $ sign (rather than quotes) to denote Eiffel entities in C inline text.

October 8, 1999

Some important changes (important not for practical programming but for theoretical consistency) to the ELKS appendix: GENERAL now deferred; ceases to be a useful practical type, replaced in this role by ANY; `as_any' introduced in GENERAL to avoid typing problems. This follows discussions with Marcel Satchell; see the Talkitover group.

October 8, 1999

Finalized (for the moment) the externals chapter by updating the Cecil part (which is now consistent with the current Cecil, see the online manual) and integrating some of the suggestions from Pierre Métras on the Talkitover group.

October 5, 1999

Fixed a mistake in the syntax for creation expressions (the create keyword, absent from informal explanations, was still present in the formal syntax). Note comment on Talkitover group about possible "gotcha".

Cleaned up the externals chapter by fixing all cross-references ... and then messed it up again by adding at the end the text of the online CECIL documentation from the ISE Web pages. I had to do that because the CECIL text in the book, straight out of the first edition, was woefully out of date. The text I added is just the plain de-htmlized form of the online manual; in the next few days I will clean it up and merge it with the rest of the chapter. So much for previously announced plans to which, however, I will return as soon as this tedious but necessary task has been completed.

October 3, 1999

Minor fixes to external interfaces chapter following comments by Marcel Satchell. Moved PDF files back from Monash to eiffel.com.

September 26, 1999

Revised chapter on Interfaces to other languages. Note that the cross-references are wrong (and incomplete) since I had to finish the writing under Windows and I manage cross-references under Unix. They should be fixed within a week. The files are still physically at Monash.

Updated changes appendix to reflect language update.

September 5, 1999

Updated Clients and Exports chapter to account for a suggestion from Marcel Satchell to add optional header comments to a New_export_item.

Updated changes appendix to reflect language update.

September 3, 1999

Updated appendix on language evolution. I know there are far more urgent things, but I had to write a paper on the topic, which I based on the original version of this appendix, so it was a trivial matter to convert this article back into a revised version of the appendix itself.

September 2, 1999

Changed section on creation expressions in creation chapter to support revised syntax specification: creation expression can now be written just {TYPE} or {TYPE .make (...), without `create' keyword (which remains legal but becomes optional). This actually makes creation expressions much nicer to use.

August 31, 1999

Chapter 8 on routines ready (usual caveats). Includes notion of pure routines.

August 30, 1999

In the next two weeks the PDF versions of reference will be physically stored at Monash. I have updated the links on this page so you don't have anything to do unless you kept links to the PDF files at eiffel.com. I will not endeavor to keep the eiffel.com versions up-to-date at all times.

August 14, 1999

Chapter 7 on clients and exports ready (usual caveats).

Updated changes appendix to reflect language update.

July 25, 1999

Chapter 5 on features ready (usual caveats); includes description of assignment procedures and assignable attributes. Did text processing cleanup for index to make sure that it uses all the proper fonts and colors.

Updated changes appendix to reflect language update.

July 18, 1999

Cleaned up "Eiffel in a nutshell" and tutorial chapters to use new creation syntax (create) keyword and big periods for calls. Didn't touch the actual contents of the chapters. There are still some uses of !! in later chapters.

July 18, 1999

Redid most of the book's figures. I only ignored a few that can easily be copy-and-pasted from others. I focused on the cosmetics of the figure, not on their semantics, so many are probably wrong.

July 17, 1999

I have started working on the figures. The class bubbles are still too big. So far, covered only the first few chapters. (Stopped in the middle of "Feature adaptation", page 300).

July 17, 1999

Found a way to make inter-document links work, but under Internet Explorer only. Too many bugs in too many products.

July 16, 1999

Inheritance chapter released.

Significant updates to changes appendix.

July 16, 1999

I am still unable to make the inter-document links work. This is a real nuisance. We have posted two queries to two different FrameMaker groups, with no result so far. I hope we find a solution.

 Short table of contents and links to chapters

(See also Full table of contents.)

EIFFEL: THE LANGUAGE, THIRD EDITION Note: this section used to include links to individual chapters. With the growth of the text, regenerating them has become a time-consuming process because they hit limitations of text processing tools. For the moment only the full text is available. Note its size (over 13 MB). Sorry for the inconvenience.        Click here for the full text.