From Newsgroup: comp.lang.tcl

TLDR: it seems to me that "trace add variable my_array {array ...}
cmdPrefix" has rather limited use compared to {read write unset}, so
I'll really appreciate any real world docs/examples on how and when I am supposed to use it (or Tcl_TraceVar+TCL_TRACE_ARRAY in C).

Background: I have a Tcl 8.6/C++ app that exposes a tree-like C++
structure as a Tcl array: $my_array(key/subkey/subkey)
Users can:
- read a value using a key
- write a value using a key
- delete a value with [unset]
- use [array names]/get/set with a glob-pattern/* to query a subset of
this structure

This is implemented only in C++ with Tcl_TraceVar.
Although back then (years ago) the array-like API seemed simple and convenient, now I can see its limitations and rather complex C++ code,
so I'm trying to refactor it by:
- moving as much logic as possible from C++ to Tcl
- making C++ a thin layer that returns an opaque handle to the structure
and implements a set of functions to work with it
- making a proper Tcl ensemble for full access to the functionality
- on top of this ensemble I'm reimplementing the "traced Tcl array" API
for backwards compatibility with 1000s LoC that my users have written
over years. So, I've moved from Tcl_TraceVar to "trace add variable
array_name {array read write unset}"

I'm done with the read/write/unset part, this was straightforward. In
many cases users know keys in advance, and then it's enough. But often
they need to discover keys dynamically using [array names] etc, and for
this I must use the "array" ops, which after some thinking and looking
around does seem like an afterthought with serious limitations. When
this trace is invoked:

1. I don't know which subcommand caused it (names/get/set/size?)
2. I don't know what the glob pattern was
3. so, it forces me to populate the whole array with keys *and* values,
which might be expensive, even if a user is asking smth like [array
names key/subkey*] which will eventually return only a small subset of keys.
4. Finally, I have no way to distinguish [info exists my_array(key)]
from a normal read trace.

I guess, I can work around these limitations with [dict get [info frame
-2] cmd], but this looks like an awful hack, and I'm not sure how
reliable it is.

So, before I dig deeper, I'm looking around for real-world examples of
how it was supposed to be used *properly*, and what pitfalls I might encounter.
BTW I did check the Ashok's book, and it doesn't go into much detail in
this aspect either (understandably so, it's very thick already)

thanks in advance!
Petro


--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.lang.tcl


Sorry if this may seem like a silly question, but what's the point of putting a trace on the reading of a variable (or an element of an array)?
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.lang.tcl

On 5/18/2026 4:35 PM, abu wrote:

Sorry if this may seem like a silly question, but what's the point of putting a trace on the reading of a variable (or an element of an array)?

Maybe used for debugging?
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.lang.tcl

abu <[email protected]d> wrote:

Sorry if this may seem like a silly question, but what's the point of putting a trace on the reading of a variable (or an element of an
array)?

One use is as a way to generate events in plain Tcl (no Tk) code.

Also keep in mind that once the plumbing is built for tracing writes, providing traces for reads is almost all but free.

--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.lang.tcl

abu <[email protected]d> posted:

Sorry if this may seem like a silly question, but what's the point of
putting a trace on the reading of a variable (or an element of an array)?

One use is to have the tracing code *update* the value of the variable
that's being read, e.g. to have it automatically track some external data.
--
Colin Macleod ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ https://cmacleod.me.uk

FEED HOUSE SAVE FEED HOUSE SAVE FEED HOUSE SAVE
GAZA GAZA GAZA GAZA GAZA GAZA GAZA GAZA GAZA
NOW! NOW! NOW! NOW! NOW! NOW! NOW! NOW! NOW!
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.lang.tcl

On 19/05/2026 10:45, Colin Macleod wrote:

abu <[email protected]d> posted:

Sorry if this may seem like a silly question, but what's the point of
putting a trace on the reading of a variable (or an element of an array)?

One use is to have the tracing code *update* the value of the variable
that's being read, e.g. to have it automatically track some external data.

- simpler alternative to callbacks: an object can notify about its
$status or $new_data or $last_error through a variable that external
code can trace with op=write and invoke a callback
- testing: e.g. a variable is a channel, so a test framework can trace
it and wait until it is initialized, and then replace the real channel
with a "mock", e.g. reflected or transformation channel
- interface to C data, as I explained in my first message

in short, traces are amazing, and you won't find them in any other language
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.lang.tcl

* Petro Kazmirchuk <[email protected]>
| On 19/05/2026 10:45, Colin Macleod wrote:
| > abu <[email protected]d> posted:
| >
| >>
| >> Sorry if this may seem like a silly question, but what's the point of
| >> putting a trace on the reading of a variable (or an element of an array)? --<snip-snip>--
| - simpler alternative to callbacks: an object can notify about its
| $status or $new_data or $last_error through a variable that external
| code can trace with op=write and invoke a callback
--<snip-snip>--

| in short, traces are amazing, and you won't find them in any other language

However, when used excessively for control flow, traces can be a
nightmare to debug or refactor, since you don't see the command flow
from the code directly.

A simple
set foo bar
can have arbitrary side effects, depending on which module has set up
the trace, whereas with callbacks you at least see in the code that
there is something going on.

If at some point, the setting of the variable is removed for any reason,
the code follows a different path, and you might not even be aware of
that. With a callback to be removed, you at least see that you need to
check where/how the callback is used.

My €0.01
R'
--- Synchronet 3.22a-Linux NewsLink 1.2