trace type (included as part of the
Tezos_lwt_result_stdlib.Lwtreslib.TRACE module is abstract in this interface but it is made concrete in the instantiated error monad (see
The idea of abstracting the trace is so that it can evolve more easily. Eventually, we can make the trace abstract in the instantiated error monad, we can have different notions of traces for the protocol and the shell, etc.
trace are the errors as received from the monad. In other words,
trace is the type of values that are seen when matching on
Error _ to, say, recover.
'error trace are kept separate (although they can be equal) to support cases such as the following:
traceare richer than
error, such as by including a timestamp, a filename, or some other such metadata.
privatetype or an
erroris the type of argument to the functions that construct the private/abstract
traceis a collection of
errorand additional functions (not required by this library) allow additional manipulation. E.g., in the case of Tezos: errors are built into traces that can be grown.
There is some leeway about what traces are, what information they carry, etc. Beyond this leeway, Lwtreslib is opinionated about traces. Specifically, Lwtreslib has a notion of sequential and parallel composition. A trace can be either of the following.
- A single-error trace, i.e., the simplest possible trace that corresponds to a simple failure/issue/unexpected behaviour in the program. See
- A sequential trace, i.e., a trace of errors where one precedes another. This is used to contextualise failures. E.g., in a high-level network handshaking function, a low-level I/O error may be built into a trace that shows how the low-level error caused a high-level issue). See
- A parallel trace, i.e., a trace of errors that happened in concurrent (or non-causally related) parts of the program. See
val make : 'error -> 'error trace
make e is a trace made of one single error.
cons e t is a trace made of the error
e composed sequentially with the trace
Typically, this is used to give context to a low-level error.
let query key = let open Lwt_result_syntax in let* c = connect_to_server () in let* r = send_query_over_connection c key in let* () = check_response r in return r let query key = let open Lwt_syntax in let+ r = query_key in match r with | Ok r -> Ok r | Error trace -> Error (cons `Query_failure trace)
val cons_list : 'error -> 'error list -> 'error trace
cons_list error errors is the sequential composition of all the errors passed as parameters. It is equivalent to folding
List.rev (error :: errors) but more efficient.
errors are separated as parameters to enforce that empty traces cannot be constructed. The recommended use is:
match a_bunch_of_errors with |  -> Ok () (* or something else depending on the context *) | error :: errors -> Error (cons_list error errors)
conp t1 t2 is a trace made of the traces
t2 composed concurrently.
conp_list trace traces is the parallel composition of all the traces passed as parameters. It is equivalent to
List.fold_left conp trace traces but more efficient.
traces are separated as parameters to enforce that empty traces cannot be constructed. The recommended use is:
match a_bunch_of_traces with |  -> Ok () (* or something else depending on the context *) | trace :: traces -> Error (conp_list trace traces)
Note that the Lwtreslib's library does not require it, but it is recommended that you also make, for your own use, a pretty-printing function as well as some introspection functions.
One possible extension can be found in
pp_print pretty-prints a trace of errors
pp_print_top pretty-prints the top errors of the trace
val fold : ( 'a -> 'error -> 'a ) -> 'a -> 'error trace -> 'a
fold f init trace traverses the trace (in an unspecified manner) so that
init is folded over each of the error within
f. Typical use is to find the worst error, to check for the presence of a given error, etc.