Hi, starting with LuaMetaTeX 2.07.01, a C library I load from Lua started to segfault and valgrind indicates that it accesses invalid memory in multiple places in the middle of Lua internal functions?! The whole thing looks almost like the Lua versions in LuaMetaTeX and in the library are incompatible, but both seem to use 5.4.0. So I asked myself if there might be changes not reflected by the version number. tl;dr: Has the embedded Lua interpreter changed in some non ABI-compatible way in LuaMetaTeX 2.07.01? (Especially on Linux, x86-64. I didn't test other platforms yet) Best regards, Marcel
Hi,
starting with LuaMetaTeX 2.07.01, a C library I load from Lua started to segfault and valgrind indicates that it accesses invalid memory in multiple places in the middle of Lua internal functions?!
The whole thing looks almost like the Lua versions in LuaMetaTeX and in the library are incompatible, but both seem to use 5.4.0. So I asked myself if there might be changes not reflected by the version number.
tl;dr: Has the embedded Lua interpreter changed in some non ABI-compatible way in LuaMetaTeX 2.07.01? (Especially on Linux, x86-64. I didn't test other platforms yet)
On 7/29/2020 6:22 AM, Marcel Fabian Krüger wrote: there have been a couple of bug fixes after the formal 5.4.0 release but afaik not in the virtual machine code (which normally is the most sensitive); i used to mark low level changes so that it got signaled (in context) but the byte code version info changed late in 5.4 dev so i no longer do that now although i occasionally have run into buglets in 5.4 code - which to me is quite natural as it concerns dev / exp versions so one gets what one downloads, i.e. one can only blame oneself for using such versions and hardly complain - lua is quite stable and reliable (given the massive amount of lua code that we use in context) ... so, normally patience - read: wait for updates - is the answer Hans ----------------------------------------------------------------- Hans Hagen | PRAGMA ADE Ridderstraat 27 | 8061 GH Hasselt | The Netherlands tel: 038 477 53 69 | www.pragma-ade.nl | www.pragma-pod.nl -----------------------------------------------------------------
there have been a couple of bug fixes after the formal 5.4.0 release but afaik not in the virtual machine code (which normally is the most sensitive); i used to mark low level changes so that it got signaled (in context) but the byte code version info changed late in 5.4 dev so i no longer do that now
although i occasionally have run into buglets in 5.4 code - which to me is quite natural as it concerns dev / exp versions so one gets what one downloads, i.e. one can only blame oneself for using such versions and hardly complain - lua is quite stable and reliable (given the massive amount of lua code that we use in context) ... so, normally patience - read: wait for updates - is the answer
Oh I see, I wasn't aware that the dev versions are used. After recompiling with the latest development code the error is gone. One other question about LuaMetaTeX 2.7.1: I noticed that the terminal_input callback is gone. Does this mean that the current behaviour of basically freezing when previously terminal input was requested is hardcoded or is some other callback invoked which can then provide some tokens? Marcel
On 7/29/2020 6:11 PM, Marcel Fabian Krüger wrote:
there have been a couple of bug fixes after the formal 5.4.0 release but afaik not in the virtual machine code (which normally is the most sensitive); i used to mark low level changes so that it got signaled (in context) but the byte code version info changed late in 5.4 dev so i no longer do that now
although i occasionally have run into buglets in 5.4 code - which to me is quite natural as it concerns dev / exp versions so one gets what one downloads, i.e. one can only blame oneself for using such versions and hardly complain - lua is quite stable and reliable (given the massive amount of lua code that we use in context) ... so, normally patience - read: wait for updates - is the answer
Oh I see, I wasn't aware that the dev versions are used. After recompiling with the latest development code the error is gone.
One other question about LuaMetaTeX 2.7.1: I noticed that the terminal_input callback is gone. Does this mean that the current behaviour of basically freezing when previously terminal input was requested is hardcoded or is some other callback invoked which can then provide some tokens? first line on console was already gone (only the filename gets pushed into the input because tex forces an input
anyway, read/write channel zero are for the terminal Hans ----------------------------------------------------------------- Hans Hagen | PRAGMA ADE Ridderstraat 27 | 8061 GH Hasselt | The Netherlands tel: 038 477 53 69 | www.pragma-ade.nl | www.pragma-pod.nl -----------------------------------------------------------------
On Wed, Jul 29, 2020 at 09:37:13PM -0700, Hans Hagen wrote:
One other question about LuaMetaTeX 2.7.1: I noticed that the terminal_input callback is gone. Does this mean that the current behaviour of basically freezing when previously terminal input was requested is hardcoded or is some other callback invoked which can then provide some tokens?
Marcel
first line on console was already gone (only the filename gets pushed into the input because tex forces an input
I think this is a misunderstanding. I wasn't referring to the first line, but to lines after the main input is fully evaluated. E.g. as a ConTeXt example \starttext \def\dosomethinguseful#1{Do something useful with #1} \dosomethinguseful{ Ups, I "forgot" to close this brace \stoptext This used to end up asking for terminal input. Especially since the terminal_input callback was available, it was possible to detect such situations and trigger an error. With the current ConTeXt code, I couldn't find a nice solution for this and when running such documents, the interpreter basically freezes, starts consuming memory and finally gets killed by the out of memory handler. Experients with token.scan_next suggest that instead of asking for input, TeX keeps seeing a special relax-like token. I was hoping that there is some callback which can be used to intercept this and either fail gracefully with an error message or to provide more useful tokens.
anyway, read/write channel zero are for the terminal
That's good to know, thank you. Marcel
participants (2)
-
Hans Hagen -
Marcel Fabian Krüger