Re: [TCLCORE] The ongoing dispute over TIP#509

[Harald O. <har...@el...>]

Thanks, TIP is online.
We would need an implementation in the main branch.

A clear warning "breaking backward compatibility of a never working 
feature" would also be great for formal reasons.

Thanks for all,
Harald

Am 13.10.2025 um 11:38 schrieb Christian Werner:
> On 10/13/2025 11:04 AM, Harald Oehlmann wrote:
> 
> Harald,
> 
>> ...
>> I will sponsor with pleasure.
> 
> fine, here we go. Please add this markdown to the TIP repo.
> 
> -----8><-----8><-----
> # TIP 734: Revert TIP 509
>          Author:         Christian Werner <und...@go...>
>          State:
>          Type:           Project
>          Vote:
>          Created:        13-Oct-2025
>          Post-History:
>          Keywords:       Tcl,threads
>          Tcl-Version:    9.1
>          Vote-Results:
>          Votes-For
>          Votes-Against:
>          Votes-Present:
>          Tcl-Branch:
> -----
> 
> # Abstract
> 
> This TIP proposes to revert TIP 509 which introduced reentrant
> mutexes on all core-supported platforms.
> 
> # Context
> 
> The original TIP 509 was invented to improve POSIX signal handling
> in multi-threaded Tcl and later overcome by TIP 511 which solved
> POSIX signal handling without requiring mutex reform.
> 
> # Rationale
> 
> Making all Tcl_Mutexes reentrant interacts bad with Tcl_Conditions
> since in order to successfully (legally) wait on a condition requires
> the associated mutex to be unlocked exactly before the condition
> is waited on and re-locked exactly afterwards. However, a deadlock
> is most likely due when the mutex was (reentrant) locked more than
> once. To overcome this situation additional code would be needed in
> the Tcl_Mutex lock and unlock operations as well as in the wait on
> condition operation to prevent from this potential deadlock situation.
> Effectively, more logic is needed around synchronization operations,
> i.e. it is not possible anymore on the theoretical ideal POSIX platform
> to just wrap the native APIs for the Tcl_Mutex and Tcl_Condition
> synchronization objects.
> 
> Moreover, currently there are zero places in the Tcl core which
> require Tcl_Mutexes to be reentrant, so this feature can be regarded
> as superfluous anyway.
> 
> # Specifications
> 
> This TIP proposes to revert all TIP 509 changes and bring the Tcl_Mutex
> system back to its 8.6 state.
> 
> # Alternatives
> 
> Should there be real demand in having reentrant mutexes in Tcl
> a better alternative would introduce a new type, e.g. Tcl_ReentrantMutex,
> which provides this feature and simultaneously prevents from the
> above mentioned potential deadlocks since it cannot be used for
> conditions.
> 
> # Copyright
> 
> This document has been placed in the public domain.
> -----8><-----8><-----
> 
> Thank you,
> Christian