IPM 0.9 dependency resolution taking a long time

[isc-shuliu]

Some packages will sporadically take a long time to install on the v1 branch which has IPM 0.9.x.

Steps to Reproduce

To reproduce this issue:

1. git clone the repo and switch to the v1 branch
2. docker compose -d --build
3. docker exec -it ipm-iris-1 bash
4. Inside the container, run iris session iris
5. In the IRIS session, run zpm "install -v demo-pandas-analytics"
6. Observe the output, for me, it keeps hanging after the following output:

Module demo-pandas-analytics was downloaded from registry (https://pm.community.intersystems.com/)
Loading from /usr/irissys/mgr/.modules/USER/demo-pandas-analytics/1.0.0/

Load started on 12/10/2024 19:47:39
Loading file /usr/irissys/mgr/.modules/USER/demo-pandas-analytics/1.0.0/module.xml as xml
Imported document: demo-pandas-analytics.ZPM
Load finished successfully.

Dependencies:
dsw 3.2.26 @ registry
mdx2json 3.2.41 @ registry
yaml-utils 0.1.4 @ registry
zpm-registry 1.3.2 @ registry

7. Press Ctrl+p followed by Ctrl + q to detach from the container.
8. Run docker compose down && docker compose up -d --build && docker exec -it ipm-iris-1 /bin/bash to start a fresh container
9. Run iris session iris to start an IRIS session
10. In the IRIS session, run the following commands to install dependencies first

• zpm "install -v dsw 3.2.26"
• zpm "install -v mdx2json 3.2.41"
• zpm "install -v yaml-utils 0.1.4"
• zpm "install -v zpm-registry 1.3.2"

11. Now zpm "install -v demo-pandas-analytics" finishes within a reasonable period of time.

[isc-shuliu]

A number of packages are influenced by this bug, including

• demo-pandas-analytics
• iris-flow
• telegram-adapter-demo
• openapi-suite
• iris-flow

[isc-tleavitt]

This is a deadlock in %SYS.NAMESPACE resulting from 6ed90b1 - the logic here is very flawed, we explicitly add mappings for everything because of course the item doesn't exist yet, we're loading it for the first time.

This needs to be fixed for 0.9 release and may impact HS adoption.

cc @isc-kiyer @isc-eneil @isc-jlechtne @isc-jili

[isc-kiyer]

@isc-tleavitt thanks for flagging that. We would not run into it for hscore since we use -DNoMapping=1 but would for other components. Also, the code should probably use %SYS.Namespace:Get*Dest methods to check if the item would be loaded into the expected database instead of already checking existence.

cc @isc-shuliu @isc-eneil @isc-jlechtne @isc-jili