Re: -no-autoloads broken in 21.2.35 --pdump

Gunnar> I think this patch is wrong. Sorry for not reviewing it Gunnar> earlier. Yeah, well, it'll get fixed before release. Thanks! Gunnar> - a variable (say, inhibit_autoloads) is declared in Gunnar> emacs.c directly in C with 'int inhibit_autoloads;' no Gunnar> DEFVAR macros involved! Huh? emacs.c, line 3393. The DEFVAR isn't an issue because of the initialization; it's a problem because it imports the variable to pdump land. This allows the pdumper to overwrite the user's command line settings. Gunnar> note it doesn't get initialised to anything. It's a global. I thought globals were guaranteed to be initialized to zero---that's what `Internals | Rules When Writing New C Code | Adding Global Lisp Variables' says. Is that incorrect? Maybe it's stylistically bad, but technically it should work, right? Gunnar> In particular all this breaks for values which get set to Gunnar> 1 by default in vars_of_emacs(), like Gunnar> inhibit_site_[lisp|modules] (at least depending on Gunnar> configure time constants). This is incorrect implementation of configure constants if the variable is settable from the command line, I believe. I should have caught and fixed that. (Et tu, Olivier! ;-) Gunnar> I assume that this last sentence is meant to refer to Gunnar> INHIBIT_SITE_LISP and _MODULES? They are always Gunnar> DEFVAR'd. It is the *initialisation* to 1 that is Gunnar> conditional. No, it's general advice to anybody who implements a command argument in the future, and means what it says. I didn't mean to say that if an arbitrary variable _is_ DEFVAR'd it _must_ be saved/restored; that is obviously false---(p)dumping couldn't work at all. I meant that command arguments that are not DEFVAR'd can't be stomped, and don't need saving. I will improve that comment immediately (and the rest of the patch pending discussion ;). It's clear I did the wrong thing in conditionalizing the saves. For variables that are settable by command line option, the saves/restores should be unconditional, and the conditional initialization should be moved to the declaration. I would argue that initialization of a variable settable by command line option must not be done in vars_of_*, and it should be done at the declaration. At least for these pdump-affected variables. Gunnar> Finally I don't see a way to affect inhibit_site_lisp by Gunnar> command line options anyway, thus making the whole Gunnar> save&restore thing overkill. That is correct. I don't remember accurately what I was thinking at the time, but I guess that I traced inhibit_autoloads carefully, and then assumed the others were all was analogous to it. That assumption is false, as you point out. -- University of Tsukuba Tennodai 1-1-1 Tsukuba 305-8573 JAPAN Institute of Policy and Planning Sciences Tel/fax: +81 (298) 53-5091 _________________ _________________ _________________ _________________ What are those straight lines for? "XEmacs rules."